Aiming at the uncertainty of the tool wear acoustic emission signal and slow convergence speed, easy to fall into local minimum value, and higher feature requirements of the neural network learning algorithm, the method for tool wear state recognition is put forward based on cloud theory and least squares support vector machine (LS-SVM). First of all, the acoustic emission signal is decomposed and reconstructed by wavelet packet, filtering out the influence of interference spectrum for calculating characteristic parameters; Secondly, reverse cloud algorithm is used for extracting the cloud characteristics parameters: expectations, entropy and hyper entropy from the reconstruction signal, and analyzed tool wear AE signal characteristics parameters of cloud and the relationship between wear and the characteristic of cloud; Finally the cloud characteristic parameters of feature vector are put into the least squares support vector machine to recognize the state of tool wear. Research results show that the extracted features could reflect the state of tool wear, cloud LS-SVM method can realize the tool wear state recognition. Compared with the traditional neural network recognition method, cloud-LS-SVM method has a higher recognition rate, and the recognition rate is 96.67%.