Blocking is an interesting phenomenon in learning. When a cue can predict the outcome effectively, it will block learning of other cues. Some versions of the error-driven learning have been incorporated into most category learning models and suggested that features useful for categorizations would block learning of other features. According to error-driven views, no matter the category has a theme or not, features can be used to classify all exemplars perfectly would block learning of other features. In contrast, some researchers indicate that different from association learning, category learning is a high level cognitive processing and people would like to learn more features than classification. When there is a theme in category, the theme knowledge helps to strengthen the activation of features and more features are learned. Are there some cognitive mechanisms in category learning in addition to error-driven mechanism? In a blocking paradigm, the category learning mechanisms have been investigated in three experiments by testing the knowledge effect in category learning. In experiment 1, rote features were defined as defining features to investigate whether the theme knowledge affected category learning. No feature learning superiority has been found in the theme knowledge condition. One possibility was that the rote defining features made participants had difficulty in getting the theme knowledge. In experiment 2, defining features were changed to knowledge features and besides defining features some features had been learned. It indicated that the theme knowledge in the category helped to learn more features and blocking effect disappeared. Experiment 3 got the same results while the defining knowledge features had been fixed on the top of exemplar. When there was a theme in category, the disappearance of blocking effect in category learning indicated that category learning was not solely driven by error-reduction. The cognitive mechanism in category learning helped to make use of theme knowledge and learn more features. A primary conclusion can be made that there are both error-driven mechanism and cognitive mechanism in category learning.