Behav Ecol Sociobiol (2000) 48:36–43
O R I G I N A L A RT I C L E
Andreas Gumbert
Color choices by bumble bees (Bombus terrestris):
innate preferences and generalization after learning
Received: 9 November 1999 / Received in revised form: 19 March 2000 / Accepted: 31 March 2000
Abstract It is usually assumed that the choice behavior
of bees for floral colors is influenced by innate prefer-
ences only for the first flower visits prior to any experi-
ence. After visits to rewarding flowers bees learn to as-
sociate their colors with a reward. This learning process
leads to an acquired preference for the trained colors that
has been believed to dominate over previous experiences
and over innate preferences. This work investigates how
bumble bees (Bombus terrestris) chose among artificial
flowers of different colors after they had been extensive-
ly trained to other colors. The bees chose novel colors
according to their similarity to the trained color if the
trained color was similar to some of the test colors. This
was true also if trained colors and test colors were well
distinguished, so their color choice reflected generaliza-
tion between colors. If the test colors were so different
from the trained color that no generalization took place,
choice behavior was not affected by the trained color and
reflected innate preferences. The differences in choice
frequencies could not be explained by physical proper-
ties of the test colors other than the dominant wave-
length, a parameter taken to reflect hue perception. Pre-
ferred dominant wavelengths correspond to those ob-
served in naive bumble bees and honeybees. Thus bum-
ble bees show innate preferences for certain colors not
only prior to color learning but also after intensive learn-
ing when choosing among very different novel colors.
Color choice among similar colors, however, is con-
trolled by generalization from the learned color.
Communicated by: R.F.A. Moritz
A. Gumbert (✉)
Institut für Biologie-Neurobiologie, Freie Universität Berlin,
Königin-Luise Str. 28–30, 14195 Berlin, Germany
Present address:
A. Gumbert, Institut für Neurobiologie,
Heinrich-Heine-Universität Düsseldorf, Universitätsstr. 1,
40225 Düsseldorf, Germany,
e-mail: andreas.gumbert@uni-duesseldorf.de,
Tel.: +49-0211-8113413, Fax: +49-0211-8111971
© Springer-Verlag 2000
Key words Color vision · Color learning ·
Generalization · Innate preferences · Bumble bee
Introduction
A bee that forages in a diverse plant community meets a
broad variety of flowers, differing in their appearance as
well as in quality and quantity of reward. How does a
bee select flowers out of this diverse food supply? An in-
dividual bee does not visit different flower species at
random but usually restricts its visits to a single or a
small number of species within a foraging bout, a behav-
ior that is generally termed flower constancy (Heinrich
1976; for reviews see Menzel 1985; Waser 1986; Chittka
et al. 1999). This foraging behavior can be interpreted to
be the outcome of an evaluation process in which a bee
constantly selects between potential food sources when
approaching flowers along its foraging path. What are
the rules that determine this evaluation process? Flowers
are detected and discriminated by bees according to spe-
cific signals such as color, size, shape, patterns, odor,
and other characteristics. Of these, color appears to be
particularly important for flower recognition (von Frisch
1914; Menzel and Backhaus 1991; Menzel and Shmida
1993). Both honeybees and bumble bees learn colors
quickly when they are rewarded according to color
choice, but bees also need to make decisions somehow
before any learning or after only a little learning. How
do innate and learned preferences interact?
The assumption that bumble bees exhibit color prefer-
ences, particularly for blue colors, dates back to the last
century and was based on observations of visitation fre-
quency to differently colored flowers under natural con-
ditions (Müller 1881). Honeybees that had gained forag-
ing experience in the field showed high choice rates for
artificial flowers of blue and/or yellow colors (reviewed
in Menzel 1985). Such observations led to the concept of
a flower syndrome for bee pollination, describing the
floral colors of melittophilous flowers as being mostly
blue or yellow (Fægri and Van der Pijl 1980). Such a

flower syndrome would arise by bilateral co-evolution,
with flowers presenting signals that their specific polli-
nators would utilize in order to identify suitable species.
However, neither the visitation rate to flowers in na-
ture nor the choice rate of experienced bees for colors
prove the existence of innate color preferences, because
such behavior may be the result of previous experience
with highly rewarding flowers in the field (Menzel
1985). Since the experience of insects previous to these
observations was not controlled, little can be said about
the innate nature of these preferences. The learning pro-
cess involved during visits to rewarding flowers is ap-
parent in experiments, where bumble bees and other bees
can be trained to collect sugar water from colored artifi-
cial flowers (e.g., Menzel 1967; Heinrich et al. 1977).
These experiments showed that bees learn to associate a
color with a reward and that different colors are acquired
differently. Violet (400–410 nm) was found to be learned
fastest, bluish-green (490 nm) slowest, and the other
colors including UV in between. Despite these subtle
differences, all colors are learned very well and control
choice behavior equally strongly after saturated learning
(5–7 visits). Since this period of learning is short com-
pared to the extensive foraging activity of an experi-
enced bee (Heinrich 1975), the majority of flower visits
are likely to be after saturated learning. Another finding
was that color learning is extremely flexible. Any new
color experience at the feeding site seems to override
previous experience, so that solely the new association
governs the choice behavior.
Since the learning process is so effective it might be
expected that choice behavior is solely determined by the
reward experience of bees at flowers. In experiments
where the amount of nectar in artificial feeders was con-
trolled, bees chose among them according to the propor-
tion of reward rather than by their color (Giurfa 1991;
Greggers and Mauelshagen 1997). Similar observations
have also been made in the field among flower varieties
differing in nectar content (Pankiew 1967).
Whether floral choices are affected by some kind of
innate behavior as well as such associative learning is
not entirely clear. The very first color choices of flower
naive honeybees and bumble bees appear to be governed
by innate preferences (Lunau 1990; Giurfa et al. 1995;
Lunau and Maier 1995). Both species show two peaks of
wavelength-dependent color choice behavior at 400–
420 nm and at 510–520 nm. The spontaneous choice
rates correlate with the acquisition rate of experienced
honeybees, in that colors that are innately preferred are
also learned fastest (Giurfa et al. 1995).
Thus it is fair to conclude that innate preferences for
colors in the first choices of the foraging career of bees
may exist, but after a few visits to a rewarding flower
such initial preferences may be subordinated to acquired
preferences and may even be deleted. Contrarily, innate
preferences may persist even in experienced bees but
may not be visible under normal foraging conditions. It
may thus be asked whether the very few visits to flowers
prior to learning are the only foraging decisions that are
37
influenced by innate preferences, and whether experi-
ence on rewarding flowers really erases these innate
preferences. Alternatively, might innate and acquired
processes also be kept apart for the control of choice be-
havior under conditions in which acquired information
does not help, e.g., when the animal is exposed to flower
colors very different from those learned?
To investigate this question, it is necessary to quantify
the color choice behavior of bees whose previous color
experience has been carefully controlled. To this end
flower-naive bumble bees (Bombus terrestris) were first
trained extensively on artificial colored flowers and sub-
sequently tested with colors that were novel to them. The
following questions were addressed:
1. Is the choice rate of new colors influenced by forag-
ing experience on different colors? In other words, do
bees simply generalize from previously learned colors
to novel ones?
2. Do experienced foragers show innate preferences for
particular new colors that are independent of the
learning process?
3. If innate preferences for novel colors appear, do they
correspond to those of naive animals?
Methods
Apparatus
Bumblebees (Bombus terrestris) were purchased from STB Con-
trol (Aarbergen, Germany) and kept in a flight cage situated in a
glass house without any contact with colored or scented nectar
sources. The animals were fed with odorless sugar solution from a
transparent acrylic feeder, and pollen. The flight cage was con-
nected to an experimental arena by a tunnel that could be selec-
tively closed (Fig. 1). In the experimental arena colored stimuli
were presented as circular disks of 9 cm diameter with a black dot
of 1 cm diameter in the center. Eight stimuli were attached to a ro-
tating plate behind a cover. Four openings in the cover allowed the
animal to see either a single colored disk and three disks with the
same color as the background (training) or four differently colored
disks at a time (test). The stimuli subtended a visual angle of 16.7°
when seen from the entrance so that the color information is per-
ceptible already when entering the arena. Honeybees perceive the
color of stimuli that subtend an angle of 15° or more (Giurfa et al.
1995). Bumblebees, having larger eyes than the honeybee, have a
higher spatial acuity (Land 1989), and are likely to have an equal
or even smaller angular threshold for color vision, so they should
perceive the stimuli as colored when entering the arena.
Training
The experiment was divided into three phases: a pretraining phase,
in which individually marked animals learned to fly into the ex-
perimental arena, a training phase, in which each individual was
trained to forage on a single color only presented on the round
plate, and a test phase in which four different unrewarding colors
were presented on the plate in the absence of the training color.
In the pretraining phase black cardboard disks with drops of
1.5 M sucrose solution were presented in the flight cage and suc-
cessively moved through the tunnel into the experimental arena.
The individuals followed the disks until they learned to fly into the
experimental arena.
In the training phase the bumblebees were rewarded on a sin-
gle disk of one of the five training colors (Tr1–5) with 2 M su-
38
Fig. 1 Experimental setup. Bees were trained to fly from the flight
cage to the experimental arena. Bees entering through the gates
were presented with a cover with four holes (side view above,
frontal view below) giving a view of four disks attached to a rotat-
ing disk (wheel) behind the cover. Depending on the position of
the rotating disk the bee could either see a single training stimulus
and three background-colored disks (training situation) or four dif-
ferent test stimuli (test situation)
crose solution. This disk had the same training color throughout
the training procedure. Henceforth I will call one flight into the
experimental arena and the subsequent return to the hive a “forag-
ing bout”. For the first five foraging bouts the disk was presented
horizontally on the ground of the experimental arena and the ani-
mals were allowed to drink ad libitum at each visit. This was done
to increase the foraging motivation by high rewards. Next, the
disk was presented vertically at one of the four possible positions
in a random succession, while at the other three visible positions
disks of the green background color were attached. In this vertical
position a small plastic cap was placed in the central black dot of
the training stimulus to hold the drop of sucrose solution. For the
following five foraging bouts the bees were allowed to drink ad
libitum. The mere motor task of keeping a grip on the vertical disk
and finding the reward in the center was difficult, and often the
animals flew up to the ceiling and sat down if they failed to find a
reward within a few seconds. In this case they were gently caught
and released in front of the four disks again until they learned to
find the reward. Later, when the animals were familiar with the
situation, the reward was reduced so that an individual had to visit
the training stimulus five times before it was satiated. In order to
change the position of the rewarding stimulus between visits the
bee was chased away from the stimulus with a piece of cardboard
of the background color so that it could not see the next position
of the rewarding stimulus before removal of the cardboard. This
procedure was repeated for six bouts with five flower visits within
each bout. Thus, at the end of the training phase each bumblebee
had made a total of 40 visits to the stimulus of the training color.
Test phase
In the test phase four unrewarded stimuli of the different test col-
ors (Te1–Te4) were presented simultaneously. Before the test
phase started, the tunnel to the experimental arena was kept closed
for 10 min after the bumblebee left the arena, in order to increase
the foraging motivation of a bee willing to return to the experi-
mental setup and, therefore, the number of choices towards the un-
rewarded test stimuli. If the bumblebees had immediate access to
the experimental arena they would spend only a few seconds in
front of the four stimulus positions and would fly off towards the
ceiling without any attempt to choose other stimuli. The gate was
opened when the animals had already entered the tunnel and
reached the end of the tunnel close to the entrance to the training
arena. Thus, it was ensured that the stimuli subtended a minimal
visual angle of 16.7° when seen by the entering bees. For 2 min
after the bee had entered the arena all significant approaches to-
wards the test stimuli, touches by the front legs, and landings on
the stimuli were recorded as choices. In the test phase bees were
not rewarded and could not leave the experimental arena but flew
constantly in front of the test stimuli.
After the test phase the rewarded training stimulus was pre-
sented again while the bees were still in the arena, and another
training phase followed in the same way as described above, but
reduced to three bouts (instead of six in the first training phase) of
five visits each. In total, the test phase was repeated three times,
with two training phases in between. Thus, each individual bum-
ble bee was observed for 2 min three times, adding up to a total of
6 min.
Statistics
As the absolute numbers of choices differed among the three test
phases and among individuals, relative choice rates of the four test
colors were calculated for each test phase and individual and arc-
sine transformed. A repeated-measures ANOVA was performed
with test phase number and training color as independent vari-
ables. After showing that test phase number did not affect relative
choice rates the effect of training color was tested with the number
of visits pooled for each individual. The LSD test for post hoc
analysis was used to compare the effects among different training
colors. The ANOVA provides the value of F, the df (first value in
parentheses), the error df (second value in parentheses) and the
value of P.
Experimental colors
Stimuli and background cover were cut from standardized color
papers of the HKS-N-series (Hostmann-Steinberg K+E Druckfar-
ben, H. Schmincke & Co., Germany). In each experiment the
same set of four test colors (Te1: HKS33n – human violet, Te2:
HKS45n – human dark blue, Te3: HKS 46n – human light blue,
Te4: HKS59n – human blue-green) was used. The training colors
(Tr1: HKS51n – human blue, Tr2: HKS54n – human dark green,
Tr3: HKS65n – human light green, Tr4: HKS6n – human orange,
Tr5: HKS10n – human red) differed in each case. The reflectance
spectra of the test colors (Fig. 2a) as well as the training colors
(Fig. 2b) and background (HKS58n – human green) were mea-
sured between 300–700 nm with a spectral flash photometer
(SR01, Gröbel UV-Elektronik, resolution 1 nm). In order to assess
color similarities among the stimuli the color-opponent-coding
(COC) model of bee color vision was used (Backhaus 1991). This
model has been shown to predict the color choice behavior as re-
lated to color similarity with a high accuracy in the honeybee
(Brandt and Vorobyev 1997), and it is assumed to apply also for
bumble bees, which are phylogenetically closely related to the
honeybee and possess a similar set of spectral photoreceptor types
(Fig. 2c, smoothed spectral sensitivity functions as revealed by
physiological measurements by Peitsch et al. 1992). The COC
model represents perceptual differences between colors by the dis-
tance of their loci in a two-dimensional color space (according to a
city-block metric). The two dimensions of the color space corre-
spond to two types of color opponent coding neurons, A and B,
the excitation of which yields the locus of a color stimulus in the
color space. The coordinates A, B of such a stimulus are obtained
as in Menzel and Backhaus (1991):
qi qi
A = ∑ ai , B = ∑ bi (1)
i =u,b,g qi + 1 i =u,b,g qi + 1

Fig. 2. Reflectance spectra of a the training colors (Tr1–5) and b
the test colors (Te1–4), measured with a flash spectrometer. Re-
flectance ranges between 0 (no reflection) and 1 (full reflection of
the illuminating light). c The spectral sensitivity of bumble bee
photoreceptors. These data were taken from electrophysiological
measurements by Peitsch et al. (1992)
where ai={–9.86, 7.70, 2.16}; bi={–5.17, 20.25, –15.08}, i=UV,
blue or green receptor, respectively, and qi is defined as the recep-
tor-specific contrast, which is determined by the relative number
of absorbed quanta q for each receptor i with respect to the back-
ground; qi was calculated as:
qi = es,i / eb,i (2)
where e is the excitation of the photoreceptor i produced by the
stimulus s or the background b. The excitation e of each type of
photoreceptor i by a colored stimulus or the background is deter-
mined as:
700 nm
ei = ∫ I (λ ) R(λ )Si (λ )dλ (3)
300 nm
where I(λ) is the spectral intensity distribution of the illuminating
light, measured in the glass house over 2 days and averaged; R(λ)
is the spectral reflectance of the stimulus or the background, re-
spectively; and Si(λ) is the spectral sensitivity of the receptor with
the index i.
The excitation of the long-wavelength receptor is of special
importance because this receptor, in addition to its contribution to
the color vision system, mediates the recognition of stimuli whose
visual angle is too small to allow their detection as colored objects
(Giurfa et al. 1996). The brightness contrast B of the stimulus s
against the background b was calculated as:
∑ es,i
i =u,b,g
B= (4)
∑ eb,i
i =u,b,g
The loci of all experimental colors in the in the COC color space
are shown in Fig. 3. The dominant wavelength of each test color
was determined by projecting the color locus from the achromatic
center of the color space onto the spectral boundary (Wyszecki
and Stiles 1982). The spectral boundary is the line connecting the
color loci of monochromatic light, which is the colored light of
39
Fig. 3 Loci of all experimental colors in the color-opponent-cod-
ing diagram (Backhaus 1991). The diagram represents the excita-
tion values in an arbitrary scale of two color-opponent-coding
channels, A and B, that correspond to color-opponent-coding neu-
rons identified in the visual pathway of the bee. Channel A re-
ceives positive input from the long-wavelength (green) photore-
ceptor type and negative inputs from the short- (UV) and medium-
(blue) wavelength photoreceptor types. Channel B receives posi-
tive input from the blue photoreceptor type and negative from the
UV and green photoreceptor types. A color distance of 1.4 units
measured in a city-block metric corresponds to 75% discriminabil-
ity between two colors. The origin represents the green back-
ground. The bold line gives the spectral boundary loci of spectral
colors in 10-nm steps, and the mixtures of 300 nm and 550 nm
marked in 10% steps. Training colors (Tr1–Tr5) are marked with
circles, test colors (Te1–Te4) are marked with squares. Dominant
wavelengths for all stimuli were obtained by projecting the color
loci onto the spectral boundary (thin lines) and the values (nm) are
displayed around the spectral boundary. No dominant wavelength
was calculated for Tr4 and Tr5, which are bee-purple colored,
which is a mixture of short- and long-wavelength light
highest spectral purity (Fig. 3). The dominant wavelength is taken
as an indication of the hue of any colored stimulus. The spectral
purity p of a color stimulus was calculated as the distance of the
color locus c from the achromatic center relative to the locus of
the corresponding dominant wavelength d:
A + Bc
p= c (5)
Ad + Bd
Spectral purity might be an indicator of the perceptual phenome-
non of saturation, but no assumptions are made here about the per-
ception of saturation or hue by bumble bees. The color distance cd
of each test color from the background was calculated as the dis-
tance from the respective color locus to the background:
cd = |Ac|+|Bc| (6)
The four test stimuli differed strongly in brightness contrast with
the background, as well as in long wavelength receptor contrast to
the background and in dominant wavelength. Differences in color
contrast with the background were less pronounced (Table 1).
40
Table 1 Physical properties of the test stimuli

Test Dominant Spectral Brightness Green Color Color distance from training colors
color wavelength purity contrast with contrast with distance from
(nm) background background background Tr1 Tr2 Tr3 Tr4 Tr5

Te1 406 0.61 3.55 1.19 5.37 4.39 6.07 9.29 9.46 11.47
Te2 411 0.46 3.17 1.51 4.77 3.10 4.78 8.95 9.12 11.13
Te3 438 0.29 2.44 2.34 4.47 1.11 3.25 8.65 8.82 10.83
Te4 480 0.67 1.14 1.83 5.10 1.42 3.88 9.29 9.45 11.46

Results
During the test phase bumblebees flew in front of the
vertical disk and showed a mean of 24.35 (SE=1.46)
choices of the test stimuli. Fewer choices were made in
the course of the three subsequent test phases but the rel-
ative choice rate did not change throughout these tests
(F(2, 88)=2.042, P=0.136). For further analysis the data
for each color were therefore pooled among the three test
phases.
The relative choice rate of new colors was significant-
ly influenced by the color of the training stimulus (cu-
mulative effect of test color and training color: (F(12,
99)=14.498, P<0.0001; Fig. 4). After training to Tr1, the
four test colors were not chosen equally (F(3, 24)=27.179,
P<0.0001; Fig. 4). The loci of all experimental colors in

Fig. 5. a Relative choice rates for stimuli of different dominant

wavelengths pooled after training to Tr3, Tr4, and Tr5 (means±
SE). b Relative choice rates for stimuli of different dominant
wavelengths by naive bumble bees, redrawn from Lunau (1990,
his Fig. 6) (means±SD). c Relative choice rates for stimuli of
different dominant wavelengths by naive honeybees after training
on uncolored gray (bold line and squares) and aluminum (thin line
and circles) stimuli, redrawn from Giurfa et al. (1995, their
Fig. 3). d Relative choice rate of honeybees trained to monochro-
matic light stimuli by one learning trial, redrawn from Menzel
(1967, his Fig. 9). A choice rate of 0.5 means a random choice be-
tween the trained and the untrained alternative color (thin line).
For the sake of comparison, only those data points from the three
publications were chosen that corresponded to the wavelength
range of 400–500 nm tested in the experiment presented here, and
the respective maxima were normalized to 1

the color space are shown in Fig. 3, and color distances

Fig. 4 Relative choice rates for the test colors Te1–Te4 after train-
ing to training colors Tr1–Tr5 (means+SE). Tests of differences in
relative choice rates were performed for each training color sepa-
rately using ANOVA for repeated measurements. Levels of signif-
icance are indicated by asterisks (*P<0.05, **P<0.01,
***P<0.001, ns P>0.05). The number of individuals used in each
group is indicated by N, and the total number of choices is indicat-
ed by n
between the training and test stimuli are listed in Table 1.
The test color most similar to the training color Tr1 was
chosen most frequently (Te3). Te4 ranked second, while
there was no difference between Te1 and Te2. In other
words, bees might have simply generalized the trained
color to the most similar ones.
A similar pattern was obtained after training to Tr2.
In this case the choice frequencies differed also signifi-
cantly (F(3, 20)=17.471, P<0.0001, Fig. 4), and test color

Te4 was preferred over the others. No significant differ-
ences were found in the choice frequencies between the
test colors Te1, Te2, and Te3. Te3 was chosen less fre-
quently than Te4, although is has a slightly shorter color
distance to Tr2 (Table 1).
After training to color Tr3, however, the choice fre-
quencies showed a different pattern: Te1 was chosen
most frequently, followed by Te2, whereas Te3 and Te4
were chosen at the same rate (F(3, 24)=16.823, P<0.0001).
The ranking of the choice frequencies does not reflect
the similarity between training and test colors as indicat-
ed by color distances (Table 1). Choice frequencies after
training to Tr4 and Tr5 also differed significantly (Tr4:
F(3, 40)=20.949, P<0.0001; Tr5: F(3, 24)=21.200, P<0.0001).
The choice frequencies obtained after training to Tr3,
Tr4, and Tr5 were not significantly different from each
other (F(6,66)=0.471, P=0.827). They could therefore be
pooled and plotted as a function of the dominant wave-
length of the test color (Fig. 5a). This function can be
compared to data from other authors who tested sponta-
neous choice rates of naive animals to colored stimuli of
different dominant wavelengths (Lunau 1990: bumble
bees, Fig. 5b; Giurfa et al. 1995: honeybees, Fig. 5c) and
data about the learning acquisition of experienced ani-
mals as measured by the number of choices to the re-
warding color after a single learning trial (Menzel 1967:
honeybees, Fig. 5d). All graphs show high choice rates
for stimuli of dominant wavelengths between 400 and
420 nm, but low choice rates between 430 and 490 nm.
Discussion
Influence of the training color on the choice rates
for test colors
Bumble bees chose differentially among the four test
colors in all experiments. The training color had a signif-
icant effect on the choice frequencies of new, unknown
colors. If the training color was very similar to the test
colors (Tr1), the choice frequencies were ranked accord-
ing to their color distance from the training color. Such
an influence of training to a particular stimulus on choice
decisions towards other, similar, stimuli is usually re-
ferred to as generalization (Pearce 1997). It is important
to note in this context that stimuli are defined by their
perceptual quality and not by their physical properties.
Thus, even if stimuli differ in physical properties such as
their reflectance spectra, they may not differ in how they
are perceived, so that a high choice rate for a test color
may by the result of imperfect discrimination between
training and test color. Likewise, stimuli differing only
marginally in their physical properties may well be dis-
criminated. Colors have, therefore, to be expressed in a
way that captures in a quantitative way the perceptual
value of colors. This requirement is met by calculating
color loci from spectral reflection functions using the
COC model of color vision (Backhaus 1991). This mod-
el also provides an estimate of perceptual distances be-
41
tween color loci. Therefore, discrimination values after
differential training can be compared with perceptual
color distances. Furthermore, it is important to note that
generalization between stimuli does not mean that the
stimuli could not be discriminated (Honig and Urcuioli
1978). Animals often judge stimuli as being similar to
each other even if they are able to discriminate them
very well, thus they generalize between stimuli. However,
generalization would be expected to be lower for stimuli
that differ more than stimuli that differ less.
In a preliminary discrimination experiment, in which
either Tr1 or Te4 was rewarded and the other was not,
both colors were found to be discriminated to a level of
75% correct choices. This level lies between perfect dis-
crimination (100%) and random choice (50%) and can
be used as a measure of discrimination threshold be-
tween the colors. The discrimination threshold would
thus lie at 1.4 units in the COC diagram. All other color
pairs except Tr1-Te3 have a greater color distance and
are expected to be well discriminated. It can therefore be
concluded that a high choice frequency between a train-
ing color and a test color with a smaller color distance
should not be the result of imperfect discrimination but
rather that of a generalization process. The high choice
rate for Te3 after training to Tr1, however, could be the
result of both colors being below the discrimination
threshold.
Te4 was also chosen most frequently after training to
Tr2, which has a color distance beyond the discrimina-
tion threshold. However, from the color distance be-
tween Tr2-Te3 as compared to Tr2-Te4 it may be expect-
ed that Te3 would be chosen even more frequently than
Te4, which was not observed. In this case it was not pos-
sible to exclude the possibility that generalization from
Tr2 to Te4 was stronger because of physical properties
other than the color distance. In terms of the dominant
wavelength Tr2 is more similar to Te4 than to Te3
(Fig. 3). It is also possible that with increasing color dis-
tance among the colors considered, the metric of the col-
or space does not sufficiently correspond with perceptual
color distances.
Innate preferences for test colors
If the color distance between the training and the test
colors was increased further, the influence of color dis-
tance on choice frequencies disappeared. However, even
if there was no influence of a dissimilar training color on
the choice frequencies of the four test colors, they were
not chosen equally frequently. After training to Tr3, Tr4,
and Tr5, which had a greater color distance from the test
colors, Te1 was always chosen more frequently than any
other, and Te2 ranked second. In these cases, generaliza-
tion cannot explain the pattern of choices observed.
Moreover, the same pattern was obtained after training
with different colors. If the differential choice rate for
the test colors under these conditions is independent of
the learning process at the training color, and does not
42
reflect generalization, then it must reflect innate color
choice behavior.
Which properties of innate choice behavior may con-
trol the choice rates observed for the test colors? In the
experiment of Lunau (1990) naive bumble bees preferred
colors of higher spectral purity. However, differences in
spectral purity cannot account for the different choice
rates in the experiments presented here. Had the animals
used this physical property of the stimuli to guide their
choices, then the predicted ranking of the test colors
would be Te4>Te1>Te2>Te3, which was not observed.
Furthermore, differences in the long-wavelength receptor
contrast to the background (predicted ranking: Te3>
Te4>Te2>Te1) and in color contrast to the background
(predicted ranking: Te1>Te2>Te4>Te3), or in brightness
contrast (predicted ranking: Te1>Te2>Te3>Te4; with
strong differences in brightness contrast between Te3
and Te4) also cannot explain the observed ranking of
color choice behavior. Long-wavelength receptor con-
trast and brightness contrast did also not affect the
choice behavior of naive animals (Lunau 1990; Giurfa et
al. 1995). Long-wavelength receptor contrast was shown
to be important in the honeybee for the detection of
small objects, but is not used for the recognition of ob-
jects subtending an angle of more than 15° (Giurfa et al.
1996, 1997). The experimental design chosen here ex-
cluded a choice for test stimuli from a long distance so
that long wavelength receptor contrast is expected to be
of little effect here. Choice frequency was correlated
with color contrast against the background in the experi-
ment of Lunau (1990) and Lunau et al. (1996), but not in
that of Giurfa et al. (1995). Color contrast with the back-
ground and spectral purity are correlated to each other.
Therefore it is not possible to relate any one of the physi-
cal properties of the colored stimuli to the response rat-
ing in the experiments by Lunau and co-workers.
Correspondence of innate color preferences to those
of naive animals
The data are in good agreement with observations of
Menzel (1967), Lunau (1990), and Giurfa et al. (1995) if
it is assumed that the choice rate is determined by color
hue. The dominant wavelength of a color is taken as a
physical correlate of hue, as is usual for human vision
(Wyszecki and Stiles 1982). The bumble bees in the ex-
periment presented here preferred Te1 (406 nm) over
Te2 (411 nm), and these two over Te3 (438 nm) and Te4
(480 nm). These findings correspond well with the re-
sults obtained from experiments performed with naive
honeybees (Giurfa et al. 1995): colors between 409 and
414 nm were chosen more frequently than colors be-
tween 434 and 483 nm (Fig. 5c). The acquisition rate of
honeybees trained with monochromatic lights shows a
similar dependence on the wavelength (Menzel 1967,
Fig. 5d). After a single training trial the relative choice
rate for stimuli of 413 nm was higher than that for longer
wavelengths followed by 428 nm, 444 nm, and 494 nm.
Both findings were interpreted as indicating a differen-
tial preparedness to learn these colors. Naive bumble
bees showed a higher choice rate for stimuli of a domi-
nant wavelength between 400–410 nm than to those be-
tween 450 and 490 nm (Lunau 1990; Fig. 5b). Unfortu-
nately, Lunau did not analyze whether this difference
was significant. He tested for a linear correlation be-
tween wavelength and choice rate and did not find a sig-
nificant result. However, a second maximum of choice
rate was present below 510 nm, which may have
obscured the existence of a correlation in his analysis.
It can be concluded that both naive honeybees and
naive bumble bees show a particular preference for
wavelengths between 400 and 420 nm, and that this pref-
erence is maintained even after extensive training to
other colors. However, experience with colors can inter-
fere with this innate preference as long as the trained
color is similar to the variety of colors that the bee has to
choose among. The conclusion that the pattern of prefer-
ences that a naive animal evinces in its first foraging
bout is maintained, and not erased by cumulative experi-
ence with other colors, is important. The pattern of pref-
erences remains inhibited, and can only be expressed
again under particular circumstances.
Taking the results presented together with those cited
above it is obvious that it is not only the very first flower
choices in the life of a bee that are governed by innate
color preferences. Bees rather appear to refer to these
preferences in situations in which their preferred flower
types are not available and they have to switch to new
species. This foraging behavior has implications for the
pollination of plant species that are visited mainly due to
their floral signals rather than due to their reward. How
much floral signals influence the visitation rate to a plant
species should depend on the abundance of the plant and
the amount of reward it offers. Abundant and highly re-
warding plants can train their pollinators, which learn to
identify the plant by its floral signals. In contrast, rare
and dispersed plants have little opportunity to train their
individual pollinators, but will rather receive a large
number of flower visits from individuals that have no
previous experience on these species. Their flower
choice behavior may be more influenced by their experi-
ences on other species or by innate preferences. A study
of flower colors in plant communities revealed that, in-
deed, rare plants depend to a higher degree on their floral
signals than abundant plants (Gumbert et al. 1999). Rare
plants showed a higher tendency to either diverge or
converge in their color signals to other plants than com-
mon plants.
Plants that do not provide a reward for their pollina-
tors are expected to depend particularly on their floral
signals. A number of cases have been described, in
which plants are pollinated by bees that search for food
but do not receive a reward when visiting the flowers
(Dafni 1984). Such cases have to be separated from
those where the plant offers no nectar but may still be re-
warding in terms of pollen. Similarly to rare plants, non-
rewarding species are also unable to train their pollina-

tors and may follow a strategy of either mimicking other
rewarding flowers or addressing innate color evaluation
systems. In the latter case, they would also obtain a high-
er visitation rate if the pollinator has visited very differ-
ent rewarding flowers before. It would be interesting to
investigate whether rare and non-rewarding flowers ad-
dress innate color preferences more strongly when they
flower in a habitat devoid of rewarding flowers with
similar colors.
Acknowledgements I want to thank the following people for
numerous helpful comments and improvements of earlier
drafts: R. Brandt, M. Giurfa, U. Greggers, N. Hempel de Ibarra, P.
Kretzschmar, J. Kunze, and R. Menzel and two anonymous refer-
ees. This work was supported by the Deutsche Forschungsgemein-
schaft. The experiments conducted for this publication comply
with current German law.
References
Backhaus W (1991) Color opponent coding in the visual system of
the honeybee. Vision Res 31:1381–1397
Brandt R, Vorobyev M (1997) Metric analysis of threshold spec-
tral sensitivity in the honeybee. Vision Res 37:425–439
Chittka L, Thomson JD, Waser NM (1999) Flower constancy, in-
sect psychology, and plant evolution. Naturwissenschaften 86:
361–377
Dafni A (1984) Mimicry and deception in pollination Annu Rev
Ecol Syst 15:259–278
Fægri K, Van der Pijl L (1980) The principles of pollination ecolo-
gy, 3rd edn. Pergamon, Oxford
Frisch K von (1914) Der Farbensinn und Formensinn der Biene.
Zool Jb Physiol 37:1–238
Giurfa M (1991) Colour generalization and choice behaviour of
the honeybee Apis mellifera ligustica. J Insect Physiol 37:
41–44
Giurfa M, Núñez J, Chittka L, Menzel R (1995) Colour preferenc-
es of flower-naive honeybees. J Comp Physiol A 177:247–259
Giurfa M, Vorobyev M, Kevan P, Menzel R (1996) Detection of
coloured stimuli by honeybees: minimum visual angles and re-
ceptor specific contrasts. J Comp Physiol A 178:699–709
Giurfa M, Vorobyev M, Brandt R, Posner B, Menzel R (1997)
Discrimination of colored stimuli by honeybees: alternative
use of achromatic and chromatic signals. J Comp Physiol A
180:235–243
Greggers U, Mauelshagen J (1997) Matching behavior of honey-
bees in a multiple-choice situation: the differential effect of
environmental stimuli on the choice process. Anim Learn
Behav 25:458–472
43
Gumbert A, Kunze K, Chittka L (1999) Floral colour diversity in
plant communities, bee colour space and a null model. Proc R
Soc Lond B 266:1711–1716
Heinrich B (1975) Energetics of pollination. Annu Rev Ecol Syst
6:139–170
Heinrich B (1976) The foraging specializations of individual bum-
blebees. Ecol Monogr 46:105–128
Heinrich B, Mudge PR, Deringis PG (1977) Laboratory analysis
of flower constancy in foraging bumblebees: Bombus tern-
arius and B. terricola. Behav Ecol Sociobiol 2:247–265
Honig WK, Urcuioli P (1978) Reizgeneralisierung und Reizkon-
trolle. In: Anonymous (ed) Die Psychologie des 20. Jahrhun-
derts. Kindler, Zürich, pp 250–305
Land MF (1989) Variations in the structure and design of com-
pound eyes. In: Stavenga DG, Hardie RC (eds) Facets of
vision. Springer, Berlin Heidelberg New York, pp 90–111
Lunau K (1990) Colour saturation triggers innate reactions to
flower signals: flower dummy experiments with bumblebees.
J Comp Physiol A 166:827–834
Lunau K, Maier EJ (1995) Innate colour preferences of flower vis-
itors. J Comp Physiol A 177:1–19
Lunau K, Wacht S, Chittka L (1996) Colour choices of naive bum-
ble bees and their implications for colour perception. J Comp
Physiol A 178:477–489
Menzel R (1967) Untersuchungen zum Erlernen von Spektralfar-
ben durch die Honigbiene (Apis mellifica). Z Vergl Physiol
56:22–62
Menzel R (1985) Learning in honeybees in an ecological and be-
havioral context. In: Hölldobler B, Lindauer M (eds) Experi-
mental behavioral ecology. Fischer, Stuttgart, pp 55–74
Menzel R, Backhaus W (1991) Colour vision in insects. In:
Gouras P (ed) Vision and visual dysfunktion. The perception
of colour. Macmillan, London, pp 262–288
Menzel R, Shmida A (1993) The ecology of flower colours and
the natural colour vision of insect pollinators: the israeli flora
as a study case. Biol Rev 68:81–120
Müller H (1881) Alpenblumen. Engelmann, Leipzig
Pankiew P (1967) Study of honeybee preference and behaviour on
normal and standardless alfaalfa flowers. J Apicult Res 6:
105–112
Pearce JM (1997) Animal learning and cognition. Psychology
Press, Hove
Peitsch D, Fietz A, Hertel H, Souza J de, Ventura DF, Menzel R
(1992) The spectral input systems of hymenopteran insects
and their receptor-based colour vision. J Comp Physiol A
170:23–40
Waser NM (1986) Flower constancy: definition, cause, and mea-
surement. Am Nat 127:593–603
Wyszecki G, Stiles WS (1982) Color science: concepts and
methods, quantitative data and formulae. Wiley, New York