In the milling process,chatter is a self-excited vibration, which has adverse effects on the quality of the workpiece. So online milling chatter detection is very important for stable and accurate machining operation. This paper proposes an in-process feature extraction method for milling chatter based on second-order synchroextracting transform and fast kutrogram. Firstly, power spectrum entropy, mean value, standard deviation and Rényi entropy are used to preliminary determine whether the collected signal is a chattering signal. Secondly, the chattering part of the collected signal is reconstructed by second-order synchroextrating transform. Thirdly, the reconstructed signal is processed by fast kutrogram to obtain kurtosis distribution. Last, the milling chatter frequency can be obtained by envelope spectral processing of the signal part with maximum kurtosis. The experiment results show that the proposed chatter diagnosis algorithms not only can detect the milling chatter online, but also can find the chatter frequency effectively.