The typical daily load curve have great significance to the load dispatching plan and operation control. Aiming at the problem that the traditional typical daily load curve selection method can not meet the demand of the current electricity market. An improved fuzzy clustering algorithm based on the combination of self-adaptive factor and probability statistics is proposed to select the typical daily load curve. The optimal clustering number is determined by using the descriptive characteristic indexes such as daily load rate and daily load fluctuation rate. The fuzzy-discrete coefficient is introduced to identify the distortion day in the sample data, and the correlation coefficient between the daily load and the monthly average load is calculated. The typical daily load curve is selected according to the correlation coefficient. The simulation results of January 2015 load data of Xinjiang Power Grid show that the proposed method can accurately select the typical daily load curve and verify the feasibility and effectiveness of the method.