The present study was aimed to explore the effect of presenting mode of different features on the acquisition of rule-based and similarity-based knowledge in category learning. There are two prominent opposing approaches on what is acquired in category learning: the rule-based approach assumes that the optimal strategy of categorization is easy to verbalize and the category is learned via hypothesis testing, whereas the similarity-based approach assumes that the optimal strategy of categorization is not easy to verbalize and the category is learned through comparing the similarity of the stimulus to the prototype. It was found that the acquisition of rule-based and similarity-based knowledge in category learning was affected by the factors such as the amount of training, working memory, and category structures. Interestingly, recently, it has been found that participants performed better in the auditory condition than in the visual condition in the rule-based task, but participants performed better in the visual condition than in the auditory condition in the information-integration task. Thus, we expected that the mode in which the features were presented might influence the acquisition of rule-based features and similarity-based features in category learning. To address this issue, a modified prototype distortion task was adopted, in which the category member included one rule-based feature and ten similarity-based features. In the auditory-visual condition, the similarity-based features were from visual stimuli that varied along 10 binary dimensions, while the rule-based feature was from auditory stimuli that different in pitch. In the visual-visual condition, the similarity-based features were identical to the auditory-visual condition, while the rule-based feature was defined by the gray scale of the background of the visual stimulus. In the visual-auditory condition, the rule-based feature was identical to the visual-visual condition, while the similarity-based features were from auditory stimuli that varied along 10 binary dimensions. In the training phase, participates were asked to indicate which category the stimulus belongs to, and then the feedback of the correctness was provided after their response. Each block included 20 trials, and there were eight blocks, for a total of 160 trials. To examine whether people learned the similarity-based or rule-based features, we adopted the ambiguous stimuli that could be classified to different categories on the basis of similarity-based or rule-based features in the testing phase. There were 40 ambiguous stimuli and 20 trained stimuli in the testing phase. The results showed that for all participants in the three conditions the accuracy gradually increased with training, indicating all of them acquired some knowledge about the category. Interestingly, the accuracy was significantly higher for participants who categorized the ambiguous stimuli on the basis of the rule-based feature, i.e., acquired the rule-based features, than those who categorized the ambiguous stimuli on the basis of the similarity-based features, i.e., acquired the similarity-based features, suggesting that the performance for rule-based learning is superior to the similarity-based learning. More importantly, the number of participants who acquired rule-based features was significantly more than that of those who acquired similarity-based features in the auditory-visual condition, but there were no significant differences in the visual-auditory and visual-visual condition. The results suggested the presenting mode of different features influenced the acquisition of rule-based features and similarity-based knowledge in category learning, and people tended to categorize stimuli on the basis of the rule-based feature when it was presented in the auditory modality.