Behavioural tests were carried out with 9 hymenopteran insect species, which ranked certain sets of coloured stimuli according to their subjective similarity to a previously memorized stimulus. Kendall's tau co-efficient is employed for the analysis of correlation between these similarity rankings and the colour distance rankings predicted by various models of neural colour computation. The models are based on the measured spectral sensitivities of photoreceptor colour types and use a variety of simple colour coding systems to derive hypothetical colour distances. The correlation between the predictions of the models and the behavioural results serves as a measure for the likelihood of existence of a colour coding system. In all species, the similarity rankings can be best explained by assuming that colour is coded on a perceptual level by two colour opponent mechanisms. Brightness differences are ignored, indicating that an intensity-coding sub-system is not used in colour discrimination by the insects investigated. The weighting factors of the colour opponent mechanisms differ between species in detail, but not in the principles involved. It is thus possible to employ a standard measure of perceptual colour distance (colour hexagon distance) to predict the capacities of colour discrimination adequately in all the tested insects.