Considering the deficiency in the multi-sensor fusion method for cutting-tool wear monitoring, a method using multi-feature fusion and least squares support vector machines based on acoustic emission signals is put forth. First, by method of empirical mode decomposition, bispectral analysis and wavelet packet analysis, the feature of sampling signals in domain of time, frequency and time-frequency is extracted to construct a multi-feature vector. Its dimension is then reduced using kernel principal component analysis. The fusion feature is generated by extracting the principal component whose cumulative contribution rate is above 85% and rejecting the redundant feature that has a lower correlation to cutting-tool wear. Finally, the fusion feature is put into least squares support vector machines. This method can effectively recognize the cutting-tool wear condition, especially in small samples, and has higher recognition rates than that of a neural network.