Damages caused by landslides have been increasing because of the greater frequency of localized heavy rain. To prevent landslide disasters more efficiently, more studies in relation to predicting the initiation areas of debris flows that cause large-scale damage are necessary. The main purpose of this study is to develop a criterion for detecting debris flow initiation areas by using an empirical method that was chosen from several types of approaches to debris flow initiation detection. In this study, ten GIS-based geomorphological datasets were obtained from slide and debris flow initiation areas located in Seoul, Gyeonggi Province and Gangwon Province. The geomorphological characteristics of slide and debris flow occurrence areas were analyzed through a comparative analysis to identify relationships between debris flow initiation and each topographic index. An initiation criterion for debris flows based on the geomorphological characteristics was suggested using an Artificial Neural Network (ANN) model combined with a modified threshold for the relationship between slope and upslope contributing area. To validate the suitability of the initiation criterion for debris flow, sequential applications of slope failure analysis and debris flow initiation analysis were conducted on a case study site to simulate the actual debris flow events. As a result of the debris flow initiation analysis, the prediction capture rate of the debris flow initiation criterion was determined to be 88.9%. Debris flow initiation areas totaling 143,300 m2 were identified among 1,935,400 m2 of slope failure areas predicted in the slope failure analysis. It is efficient to apply the debris flow initiation criterion in the debris flow simulation because slope failure areas are only partly transformed into debris flows.