Abstract: A new method which uses LVQ neural network in the automatic classification of surface defects for cold rolled strips was presented. The problems of long time and low accuracy in the classification of multi-defect pattern types with some traditional classification algorithms were resolved. Tested by 14 main defect types collected from online data, the results demonstrated that the method of surface defects for cold rolled strips based on LVQ neural network spent little time during training and classifying, and its accuracy could be assured on the recognition process of multi-defect pattern types.