Collaboration is an indispensable activity in social groups, while category learning, as the basis of cognitive activity of understanding the world, is also inseparable from cooperation with others in reality. Hence the study of collaborative category learning is of greatly practical significance. According to a current theory of category learning——the COVIS model, there are at least two separate learning systems in category learning: one is the verbal system which is affected by working memory and executive attention, and the other is the implicit system which is separated from verbal system and will not be affected by executive function. We hypothesized that a collaborative approach based on verbal discussion and interaction would have different patterns of influence on these two kinds of learning systems. This study used the classic experimental material—Gabor for category learning. In order to explore whether difficulty affected the effect of collaboration on information integration learning, we set two decision-making difficulty levels: high and low. This study included 3 experiments of 2 (participant type: collaborative and individual group) x 2 (category structure: rule-based and information-integration) complete between-subjects design. Moreover, conjunctive rule was used in experiments instead of single-dimensional rule to control the interference of unequal difficulty between rule-based category structure and information-integration category structure on the learning effect. Experiment 1 was a low difficulty classification task, while Experiment 2a and 2b were in high difficulty. The Experiment 2a had the same procedure as the Experiment 1 where all the participants would accept an individual test after learning task. In Experiment 2b, a relearning stage where all the participants would study individually, was added on the basis of Experiment 2a so as to further explore the learning effect. An analysis of variance (ANOVA) was used to analyze the accuracy of category learning in all experiments. The results showed that in Experiment 1 with low decision-making difficulty, collaboration significantly reduced rule-based learning performance (p<0.05), but did not significantly affect information integration learning. In Experiment 2a with high decision-making difficulty, collaboration showed no significant impact on learning. However, in the further exploration in relearning stage of Experiment 2b, the difference of accuracy rate between relearning stage and first learning stage was adopted as a more sensitive measurement index. The results indicated that collaboration would interfere with both two kinds of learning under the high decision-making difficulty (p<0.05). Combining the results of Experiment 2a and Experiment 2b, it was found that collaboration significantly inhibited rule-based and information-integration learning in high decision-making difficulty. Besides, participants’ metacognition were measured in the experiments, the results illustrated that the level of metacognition in rule-based category learning was higher than that of information integration category learning (p<0.05). This study reveals the role of collaboration in perceptual category learning. The findings show that collaboration has different influence patterns on two separate learning systems. Regardless of the difficulty of learning task, the performance of the rule-based category learning will decline under the condition of collaboration comparing to learning individually. As for the implicit learning system, the anti-interference of information integration learning is only shown in low decision-making difficulty, while in high difficulty, information integration learning is also inhibited by collaboration.