Cutting chatter has a tremendous impact on the machining steadiness of machine tools and seriously restricts the machining efficiency. In order to realize the on-line monitoring and especially forecasting of cutting chatter, a method for chatter feature extraction based on mean square frequency (MSF) is proposed. By analyzing the cutting force signal in frequency domain, the ratio of MSF is calculated as a feature based on chatter sensitive frequency band extracted by wavelet packet transform. The results display that the presented method is able to identify the cutting chatter in advance effectively and realize chatter early warning, which provides conditions for further chatter suppression.