Radar micro-Doppler signatures are of great potential for identifying properties of unknown targets. An effective tool to extract information from the signatures is time-frequency analysis, based on which target identification and object recognition can be extended. In this paper, a method has been proposed for feature extraction and selection from simulated time-frequency distribution of micro-Doppler dynamics. Experimental results have shown that a highly discriminative feature set can be established by using this method. With this feature set, high classification performances both in training and testing stages for different classifiers have been achieved.