The distribution of difference feature frequency is crucial for establishing a multi-attribute fusion validity distribution synthesis of difference features of bimodal infrared images. To construct different feature frequency distributions of bimodal infrared images, a method of constructing a difference feature frequency distribution based on the K nearest neighbor(KNN) probability density estimation is proposed. The cumulative distribution function is used to obtain the true sequence value of the difference feature frequency; subsequently, the similarity measure of the statistically significant frequency sequence value and the real sequence value in the constructed frequency distribution of the difference feature are calculated. Experimental results show that non-parametric probability density estimation can be applied to the frequency distribution of difference features. The proposed method can accurately construct the frequency distribution of difference features compared with the MISE optimal bandwidth Gaussian kernel density estimation.