Abstract: This paper describes a 2D bar code extraction approach that is capable of processing pictures of 2D bar codes on metal parts with uneven distribution of light intensity,point spread,low contrast and pollution interference.The algorithm uses kurtosis value-sorting and module area fine tuning to locate each module's position which is detected from coarse-level to fine-level,then the final extraction data of 2D bar code is obtained by a genetic algorithm based on the original gray-scale image.Compared with traditional methods,the proposed algorithm has higher reliability for 2D bar code extraction on complex metal parts.