The invention relates to a collapse stack data-based software defect code locating method. Historical collapse stack data are combined with historical code maintenance records; by a machine learning technique, a defect code prediction model is established; and various final target collapse code units including defects in software are predicted, so that the code units including the defects in the software are rapidly located; evaluations are carried out on the final target collapse code units; and the work efficiency of restoring software codes is effectively improved. The collapse stack data-based software defect code locating method is simple in design method and suitable for different types of programming languages and different scales of computer software, and has scalability and adaptability; a parallel algorithm is convenient to use; a lot of collapse stack data can be rapidly and effectively analyzed and processed; and the software defect locating efficiency is improved; and the collapse stack data-based software defect code locating method can be applied to testing and maintaining work of medium-to-large software systems.