Abstract: To consider fuzziness in river runoff, an improved method for annual runoff classification and a fuzzy classification criteria of runoff are presented by analyzing the existing classification approach based on set pair analysis. We adopted a contribution weighting factor to describe both the annual amount of runoff and its features of distribution in a year, and obtained a comprehensive membership degree of annual runoff for each class. Then, we evaluated and classified the annual runoff level relative to high and low flows using a set pair analysis method. Application to the runoff analysis of the Xiluodu station at the Jinsha River shows that this method is more effective in runoff level classification than approaches based on set pair analysis.