New tool wear state recognition method based on hyper-sphere support vector machines was proposed. The correlation between the tool wear loss and the features acquired from cutting force and vibration signals of different wear states was analyzed. The mean value, mean square root, the energy and approximate entropy of wavelet coefficient were calculated and integrated as the feature vectors. Ultimately, in order to realize recognition of different wear states, hyper-sphere support vector machines (SVMs) algorithm was adopted as classifier. The results show that hyper-sphere SVMs are with excellent study ability, generalization ability and of high recognized precision with small training samples.