We model two patterns related to the acquisition of color terms in Russian and English: children produce overextension errors for some colors but not others, and language-specific distinctions affect color discrimination in a non-linguistic task. Both effects, as well as a reasonable convergence with adult linguistic behavior, are shown by a Self-Organizing Map trained on naturalistic input. We investigate the effect of different ways of representing colors, i.e., as perceptual features or in terms of the cognitive biases on categorization extracted from crosslinguistic color naming data. We also consider the influence of color term frequency. Our results suggest effects of all three of term frequency, cognitive biases, and perceptual features.