Abstract: The evaluation of texture similarity is very important in texture synthesis and content-based image retrieval systems. In these systems, texture features corresponding to human perceptions are more suitable than those features without specific correspondence to human perceptions. In this paper, 19 texture features corresponding to human perceptions are used to evaluate the similarity of textures. These 19 features and the corresponding estimation method are proposed based on work by Tamura, Amadasun and Haralick et al. The similarity between different textures is defined by the Euclid distance of their feature vectors. Principle component analysis PCA is used to reduce the number of features from 19 to 6 while keeping about 85% capability . Experiments show that the similarity criteria based on these features works well on large texture image bases.