Abstract. In this paper, we propose a novel method for generic object recognition by using higher-order local auto-correlations on probability images. The proposed method is an extension of bag-of-features approach to posterior probability images. Standard bag-of-features is approximately thought as sum of posterior probabilities on probability images, and spatial co-occurrences of posterior probability are not utilized. Thus, its descriptive ability is limited. However, using local auto-correlations of probability images, the proposed method extracts richer information than the standard bag-of-features. Experimental results show the proposed method is enable to have higher classification performances than the standard bag-of-features.