Most models indicate that category learning is a kind of associated learning driven by reducing categorization errors. According to this kind of accounts, people learn features useful for categorization solely. In contrast, some researchers indicate that different from association learning, category learning is a high level cognitive learning, and people like to learn more features than classification. A dual-processing account was advanced to describe category learning mechanism, which indicated there were both an error-driven associative mechanism and a cognitive mechanism in category learning. Three experiments inspected the dual-processing account in a blocking paradigm. According to the dual-processing account, features that were useful for categorization would block some other features learning. In experiment 1, features were presented randomly to investigate the blocking in category learning. The results showed that some nondefining features had been learned; even they were blocked by pre-learning of defining features, which supported the dual-processing account of category learning. Defining features were defined on the top of exemplars in Experiment 2 to prevent participants learning nondefining features by searching for defining features. Consistent with Experiment 1, the results supported the dual-processing account too. Different with the sentence described categories in Experiment 1 & 2, figure categories were adopted in Experiment 3 and similar results were found. In all three experiments, defining features blocked some nondefining features learning. The dual-processing account has been verified in different feature presentations and category materials.