In the tool wear condition monitoring, a large number of features reflecting different tool wear states can be extracted. Based on the features of state recognition based on Neural Network unable to remove redundant, the problems of long training time and low accuracy were caused. A tool wear condition monitoring method based on rough set and BP Neural Network was presented by aimed at these problems. By using rough set to reduce attributes, to remove redundant information, and then optimize features were optimized. Moreover the input data of neural network were reduced, and then the training time of neural network was shortened and recognition accuracy was improved. The effectiveness of this method has been proved through the analysis of the tool data from practical monitoring.