Cryocooler plays an extremely important role in the field of infrared remote sensing. The normal operation and performance of the detector will be affected if the cryocooler breaks down. A new intelligent fault diagnosis method for cryocooler has been proposed based on wavelet packet transform, genetic algorithm and SVM for rubbing fault. First, wavelet transform is applied to the vibration signal, and the vibration signal is extracted in time domain. The evaluation factors of the combined feature set are calculated by using the distance evaluation technique, and the corresponding sensitive features are selected. Then, the parameters of SVM are optimized by genetic algorithm. Finally, the selected sensitive features are input into the optimized SVM to identify different machine operation states automatically. The effectiveness of the method is verified by the fault simulation test of the cryocooler. Experimental results show that this method can identify and locate the cryocooler rubbing fault accurately, and the accuracy is 95%.