The compressor is the core component of refrigeration equipment. This paper proposeed a method of anomaly vibration detection of refrigeration compressor based on autoencoder technology in order to solve the quality control problems in the production process of refrigeration compressor. The principle and process of anomaly vibration detection were given, based on autoencoder and the acquisition method of compressor vibration signal sample. The relationship was studied between the mean square error of the input and output of the autoencoder and the distribution of the signal samples. By means of the autoencoder model, the decision fundament of the abnormal vibration of the compressor was given. From two aspects of the autoencoder parameters and the number of training samples, this paper discussed the influence of three kind of autoencoder parameters including calculation iteration number, hidden layer number and training sample number on the detection accuracy. The study finds that the number of iterations and hidden layers has an impact on the detection accuracy, but no obvious law; the number of the training samples has a significant impact on the detection accuracy. It is an effective method to improve the accuracy of autoencoder to increase the number of training samples. The autoencoder can be used to detect anomaly vibration of refrigeration compressor.