Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.


Bees trade off foraging speed for accuracy


Bees have an impressive cognitive capacity1,2,3,4, but the strategies used by individuals in solving foraging tasks have been largely unexplored. Here we test bumblebees (Bombus terrestris) in a colour-discrimination task on a virtual flower meadow and find that some bees consistently make rapid choices but with low precision, whereas other bees are slower but highly accurate. Moreover, each bee will sacrifice speed in favour of accuracy when errors are penalized instead of just being unrewarded. To our knowledge, bees are the first example of an insect to show between-individual and within-individual speed– accuracy trade-offs.


Psychophysicists studying stimulus discrimination in animals have been mainly concerned with the accuracy of discrimination, not with its speed. But in humans (see, for example, ref. 5) there is a tight relationship between the two. We therefore investigated how bumblebees might achieve a compromise between speed and accuracy while foraging from a 'virtual' meadow.

A nest box was connected to a flight arena (100 cm × 70 cm × 70 cm); one of the walls (70 cm × 70 cm) was a translucent Plexiglas screen. Virtual 'flowers' (coloured circles of diameter 25 mm) were projected onto the screen by a data projector controlled by a PC and Java software. The screen contained 46 holes, each 5 mm in diameter, arranged in a hexagonal pattern; the distance between neighbouring holes was 10 cm. Sucrose and other solutions could be applied from behind the screen with a micropipette.

Virtual flowers were projected onto 8 of the 46 possible locations on the screen in such a way that one hole in the screen formed the centre of each flower (Fig. 1). Four of the virtual flowers ('targets') were rewarding with 10 µl sucrose solution (2 M). These were coloured blue; the colour was adjusted to R = 0, G = 0, B = 255 in the eight-bit RGB (red–green–blue) colour model. Four other similarly coloured virtual flowers acted as 'distractors' (unrewarding virtual flowers: R = 0, G = 70, B = 255). Distractor flowers were charged with a droplet of water.

Figure 1: Bumblebees can choose wisely or rapidly, but not both at once.

a, Interindividual correlation between response time and accuracy of bees discriminating between two virtual flower types. Each symbol denotes the average performance of one individual bee under one experimental condition. When targets were rewarded with sucrose solution and distractors contained no reward (plain water) (blue symbols and black regression line), bees investing more time made more accurate choices. When distractors were penalized with bitter quinine solution (red symbols and orange regression line), all bees improved their accuracy. Blue arrows link the average values for individual bees under the two experimental conditions. b, A blue virtual flower with a bumblebee imbibing sucrose solution from a Plexiglas platform.

Flower locations were randomized every hour during training, and between individual foraging bouts during tests. After two days of training, bees were tested individually for three consecutive foraging bouts. Choice time was assessed as flight time between flowers; a decision was recorded when a bee made contact with the landing platform.

The average percentage of correct choices was 62 ± 11.2% (mean ± s.d.), and bees, as a group, behaved significantly differently from a random choice condition (χ2 = 8.73; d.f. = 1; P = 0.031). Between individuals, there was a strong correlation between decision time and accuracy (rs = 0.963; n = 10; P = 0.00007; Fig. 1a). The more time that an individual bee invested, the more accurate were its choices, whereas bees that made rapid choices were more error-prone. However, when errors go unrewarded, the cost of visiting the wrong flower type is comparatively low. It is not clear whether low accuracy actually reflects the limits of discrimination1,6,7.

We therefore introduced higher costs for making errors by penalizing incorrect choices with aversive quinine solution. Distractor flowers each bore a 10-µl droplet of 0.12% quinine hemisulphate salt in water. After a full day of training, bees were again tested individually for three foraging bouts. Under these conditions, bees improved their accuracy significantly to 83% (z = 2.84; n = 10; P = 0.004; sign test) at the expense of longer response times (z = 2.21; n = 10; P = 0.027). Between bees, the correlation between time and accuracy remained significant (rs = 0.723; n = 10; P = 0.018). There was also a correlation between performance of bees in the two experiments, in terms of both accuracy (rs = 0.951; n = 10; P = 0.00023) and speed (rs = 0.699; n = 10; P = 0.024).

These results show that fast and error-prone bees in the first experiment remained fast and error-prone in the second experiment, whereas slower bees were consistently more accurate. The improvement in performance was not simply an effect of prolonged training: when the quinine penalties were removed, accuracy fell to the same level as in the first experiment (average 61.4%).

We show that, as in humans8, accuracy of choice in bees depends on how much time is allocated to solving the task. Thus, whenever accuracy is quantified in discrimination tests on animals, response time should also be measured9, and the possibility of speed– accuracy trade-offs evaluated. Even individual insects vary in their reluctance to make errors.


  1. 1

    Menzel, R. in Neurobiology of Comparative Cognition (eds Kesner, R. P. & Olten, D. S.) 237–292 (Erlbaum, Hillsdale, New Jersey, 1990).

    Google Scholar 

  2. 2

    Chittka, L. & Geiger, K. Anim. Behav. 49, 159–164 (1995).

    Article  Google Scholar 

  3. 3

    Giurfa, M., Zhang, S., Jenett, A., Menzel, R. & Srinivasan, M. V. Nature 410, 930–933 (2001).

    ADS  CAS  Article  Google Scholar 

  4. 4

    Dukas, R. & Waser, N. M. Anim. Behav. 48, 1001–1006 (1994).

    Article  Google Scholar 

  5. 5

    Rival, C., Oliver, I. & Ceyte, H. Neurosci. Lett. 336, 65–69 (2002).

    Article  Google Scholar 

  6. 6

    Smith, B. H., Abramson, C. I. & Tobin, T. R. J. Comp. Psychol. 105, 345–356 (1991).

    CAS  Article  Google Scholar 

  7. 7

    Maleszka, R., Helliwell, P. & Kucharski R. Behav. Brain. Res. 115, 49–53 (2000).

    CAS  Article  Google Scholar 

  8. 8

    Zenger, B. & Fahle, M. J. Exp. Psychol. 23, 1783–1791 (1997).

    CAS  Google Scholar 

  9. 9

    Thomson, J. D. & Chittka, L. in Cognitive Ecology of Pollination (eds Chittka, L. & Thomson, J. D.) 191–213 (Cambridge Univ. Press, Cambridge, 2001).

    Book  Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Lars Chittka.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Chittka, L., Dyer, A., Bock, F. et al. Bees trade off foraging speed for accuracy. Nature 424, 388 (2003).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing