Abstract: The flash flood hazard zoning is conducted to prevent flood disasters and manage their risks effectively in Yunnan Province. Herein, we employed a self-organizing mapping method to conduct second-order clustering, utilizing five distinct techniques, namely, Ward, PAM, CLARA, K-means, and HK-means. The evaluation of clustering schemes was performed, using the Davies-Bouldin index (DBI), Silhouette coefficient (SC), and Calinski-Harabaz index (CH). Subsequently, the optimal number of zoning units was determined through the utilization of the coefficient of variation and the first-order difference of variance coefficients. The results reveal that: ① The self organizing map (SOM)+ clustering LARge applications (CLARA) method demonstrates best effectiveness according to the clustering validity test, with a DBI value of 1.0, SC value of 0.9, and CH value of 0.3334. The optimal number of clusters for flash flood hazards in Yunnan Province is determined to be 5, displaying characteristics of spatial separation among categories and similarity in hazard attributes. ② By analyzing the variation of coefficient of variation (CV) value and identifying the lowest first-order difference (FOD) value, the optimal zoning units are found out to be 16. These zoning units exhibit the characteristics of their forms similar to geomorphological units, of their numbers equal to administrative units, and of similar internal-disaster-occurrence mechanisms. ③ Through visualizing comparison of flash flood sites, precipitation and elevation geomorphology, geodetector quantitative analysis, the result shows high intra-zoning consistency and high inter-zoning heterogeneity within the 16 zones.