A method of fault diagnosis based on image analysis was proposed for the classification of faults in printing unit of printing machine. By printing specific images, the printing images of the printing unit under both normal and abnormal states were obtained, and multiple feature sets of image characterizing the printing unit state were established from the aspects of the cover rate of printing dots, gray feature and texture feature. After that, a fault recognition net was built through SVM. The correlation between the image feature sets and six kinds of faults in the printing unit was analyzed with multivariate statistical methods. The faults were correctly classified based on the image features. Verification of actual fault diagnosis tests showed that the correct rate of the proposed method reached over 90% . Image feature sets had a high classification and characterization capacity for the faults of printing machine, and provided a new method and theory for the maintenance of printing machine.