Objectively, user evaluation criteria are determined by subjective consciousness. Different evaluation criteria between users result in that the scores of multiple users for the same service are incomparable. Service recommendations that do not consider the incomparability of different user ratings cannot meet user personal preferences and real needs. Therefore, we propose an online service recommendation method for inconsistent user evaluation criteria. The method calculates online service recommendation results for users by considering the user’s preference relationship with the online service when user preferences are inconsistent. Firstly, based on the userservice scoring matrix, the user's preference relationship with the service is established. Secondly, the similarity between users is calculated according to the preference relationship. Thirdly, based on user similarity, the user's unscoring service is scored and predicted. Finally, the ranking results of the predicted scores are used as the recommendation results. In the experiments, we compare the method with the classical collaborative filtering recommendation method to verify its effectiveness. Experimental results show that the recommendation results obtained by the proposed method can meet the service preferences of most users, and at the same time obtain better accuracy than the classic collaborative filtering recommendation method.