An important reason that caused mechanical equipment failure is the rolling bearing fault. In order to improve the correct diagnosis rate of rolling bearing and recognize different faults effectively, a novel method of fault pattern recognition based on improved KNNC (K-nearest neighbor classifier) is presented. Firstly, the vibration feature index of training samples and test sample are calculated separately. So, the feature set of samples is constructed entirely. To accelerate classification speed of KNNC and eliminate the influence of bad samples, K-means clustering algorithm is used to optimize the training samples, and the obtained clustering centers are regarded as a new training set. At last, the pattern recognition can be realized by KNNC according to the new training set. Application in Rolling bearing experiment shows that the improved method can effectively and quickly separate the 4 different kinds of bearing fault pattern, with higher recognition accuracy.