In the fault diagnosis of rolling bearing, the vibration signals, which are collected from the field test, are often more complex because they unavoidably contain various noises and measurement errors, so ‘outliers’ may occur in the features extracted from the collected vibration signals. Aiming at the above problems, the agent discriminate model based optimization weighted (ADMOW) method is proposed. By using the entropy weight method (EWM), the entropy weights of the sample features are calculated first, and the features are then weighted to weaken the influence of ‘outliers’ on the modeling. Secondly, the particle swarm optimization (PSO) algorithm is used to optimize the parameters of the model, and a more accurate classification model is obtained. Eventually, ADMOW is applied to recognize defaults of rolling bearings. The test results indicate that by comparing several pattern recognition methods, the proposed method can effectively weaken the influence of ‘outliers’ and improve the recognition rate and the recognition accuracy.