Cigarette ash is one of the important material evidences in forensic science research. In this work, using the chemometric method, the energy dispersive X-ray fluorescence spectrometer (XRF) was used to inspect 42 cigarette ash samples, and the rapid non-destructive detection of cigarette ash was realized. First, after standardizing the XRF data in the early stage, the value of K was determined by the sum of square errors within the group. At the same time, K-means clustering method was used to perform preliminary clustering of the samples and explain the information of the samples and chemical components reflected by the clustering results. Then, the clustering results were verified by principal component analysis and single factor analysis of variance. Finally, the discriminant function model was established by using discriminant analysis. A classification model based on K-means clustering method was established. The results show that the K-means clustering using sum of square errors within groups is accurate and reliable. The clustering results are also well verified in principal component analysis. Different categories are well differentiated. Each element in single factor analysis of variance has good significance. Discriminant analysis shows that the accuracy of the classification model reaches 100%. This classification model is simple, fast, accurate, and reliable, which provides a reference for the public security grassroots to handle cases.