The category adjustment model of Huttenlocher, Hedges, and colleagues explains category effects on memory or percep-tion in terms of probabilistic inference. This model has been shown to account for category effects in color cognitionacross several languages, suggesting that effects of language on color cognition reflect standard principles of inferenceunder uncertainty. Previously unexamined is whether the same model can illuminate an influential intuition advanced byKay and Kempton: that language is likely to affect cognition primarily when purely perceptual discrimination of stimuli isdifficult because the stimuli are similar. Recent data by Welch et al. support this intuition. Here, we show that the categoryadjustment model accounts for these new data as well, strengthening the case for viewing category effects of language oncognition through the lens of probabilistic inference.