Abstract: Landslide hazard assessment is an important part of landslide risk assessment, which is of great significance to landslide prediction and prevention. Analytic Hierarchy Process（AHP）, expert evaluation, fuzzy comprehensive evaluation and other methods were often used in traditional landslide hazard evaluation to calculate the importance of inter-index, which were subjective and complicated. This paper introduced an improved model of mutation theory, which overcame the limitation of traditional methods and achieved higher evaluation accuracy. Firstly, according to field investigation and previous studies, 12 factors including slope, slope direction, elevation, plane curvature, profile curvature, distance from river, stratigraphic lithology, land use type, distance from fault, vegetation coverage rate, 24 h rainfall and human engineering activities were selected as influencing factors of landslide risk assessment, and the relative importance of indicators was determined by entropy weight method, and the landslide risk assessment system was established. Then the index was standardized and normalized, and the total mutation result was calculated. Finally, the fitting function was used to transform the total catastrophe result, and a new criterion of landslide risk assessment was obtained. Taking 20 landslides in Ya’an city as an example, the results showed that the accuracy of the evaluation results obtained by the catastrophe theory was 90%, and the improved evaluation results were more intuitive and accurate