Abstract: In order to avoid the condition that most of the segmentation based recognition methods have relied too much on the quality of image segments, a new object recognition method is proposed based on Multi-Level- probabilistic Latent Semantic Analysis (ML-pLSA) object recognition algorithm. Firstly, multiple segmentations at different levels are computed for each image, and then object classes on each segment region are estimated by using pLSA and Bag-Of-Words (BOW). The final results are obtained by fusing estimation results at multiple levels. The proposed algorithm is evaluated on Graz-02 dataset, a challenging dataset that contains large changes in object scale, object viewpoint and illumination condition. The experiment results demonstrate that the proposed method performs better than traditional object recognition methods in both accuracy and robustness.