A method for realizing tool wear condition monitoring using multi-feature of the cutting sound is presented. Based on empirical mode decomposition and Hilbert transformation theories, the cutting sound signal is analyzed. The energies of intrinsic modes and Hilbert spectrum in different frequency ranges are extracted as candidatefeatures of the monitoring signal. To solve the feature selection problem, the support vector machine is selected as the classifier, and the multiple population genetic algorithm is used to optimize its input features. Then, the interference features are eliminated from the candidate features. After the classifier parameters are also optimized with the multiple population genetic algorithm, the test samples are classified with the optimized classifier, and the performances of the classifiers before and after optimization are compared. The results show that the performance of the optimized classifier is significantly improved, and the method can be used effectively for identification of the tool wear condition