The fault diagnosis technology of reciprocating compressor can provide an effective guarantee for industrial production. Due to the low recognition accuracy of the traditional method, a fault diagnosis method of reciprocating compressor based on the time-frequency image GLCM-HOG features fusion of vibration signal is proposed. Firstly, the vibration signals of reciprocating compressor are processed by wavelet transform to generate time-frequency images. Secondly, the gray co-occurrence matrix (GLCM) features and directional gradient histogram (HOG) features are respectively extracted from the time-frequency images and fused. Finally, the GLCM-HOG features are input into support vector machine (SVM) to determine the state of the reciprocating compressor. The experimental results show that the accuracy rate can reach 92.33%, and the proposed method can accurately realize fault diagnosis for reciprocating compressor.