We study the problem of category identification, which involves making inferences about category membership (e.g., a `cat') given a set of features (e.g., has a tail, has four legs). We note that this problem is closely related to classification problems in machine learning, for which standard methods exist, and new methods continue to be developed. Using a large database of associations of features to animal stimuli, made by different people, we test several standard benchmark methods, including nearest neighbor, decision tree, and logistic regression methods. We also apply a new classifier, developed for image processing, which we call Sparse Instance Representation. We show that it is the best-performed, especially when constrained in a novel psychologically interpretable way. We conclude that our results argue for sparse exemplar-based representations of category structures.