A method of chatter forecast is studied by the application of continuous wavelet feature vector and support vector machine (SVM) for ball milling system. This method is based on continuous wavelet transformation to extract feature vector of milling vibration signal and multi-class spherical support vector machine is used to classify three classification and recognition such as normal milling state, chatter gestation state and chatter outbreak of state, which predicts the chatter outbreak by making a recognition of chatter gestation state. As experimental results show, there is a good ability to identify and forecast chatter in recognition and predict of milling vibration, which uses continuous wavelet feature vector and multiclass spherical SVM classifier to deal with milling vibration signal, recognition rate of chatter gestation state reaches 95.0%, then chatter outbreak state recognition rate is 97.5%.