Children acquiring languages with noun classes (grammatical gender) have ample statistical information available that characterizes the distribution of nouns into these classes, but their use of this information to classify novel nouns differs from the predictions made by an optimal Bayesian classifier. We propose three models that introduce uncertainty into the optimal Bayesian classifier and find that all three provide ways to account for the difference between children’s behavior and the optimal classifier. These results suggest that children may be classifying optimally with respect to a distribution that doesn’t match the surface distribution of these statistical features.