Prioritization of flood source areas (FSAs) is of paramount importance in flood management to adopt proportional measures within a watershed. Unit Flood Response (UFR) approach has been proposed to identify FSAs at subwatershed and/or cell scale. In this study, a distributed modified Clark (ModClark) model coupled with Muskingum flow routing method was used for hydrological simulations. Furthermore, SOMFCM clustering techniques involving Self-Organizing Feature Maps (SOFM) and Fuzzy C-Means algorithm (FCM) were used to identify Hydrologic Homogenous Regions (HHRs). The case studies were two semi-arid watersheds including Tangrah in northeastern Iran and eastern part of Walnut Gulch Experimental Watershed (WGEW) in Arizona. DEM-derived geomorphological and hydrological features were entered into Factor Analysis (FA) to determine the most effective variables in runoff generation. The optimum SOMFCM resulted in clustered HHRs map which was generally similar to that of the UFR-delineated FSAs at cell scale, especially in cases of maximum flood index values for both watersheds. Although clustering techniques, such as SOMFCM, cannot directly provide a map of FSAs involving absolute values of flood index, most dominant watershed physical features may be used to identify the most critical, or effective FSAs through clustered HHRs. Application of SOMFCM in two semi-arid watersheds demonstrated that SOMFCM provides a simple and useful tool to qualitatively identify the ranking of FSAs across a watershed. Therefore, the clustered HHRs involving higher ranks of FSAs that represent the most flood active regions, are expected to assist policymakers for effective management of floods.