Abstract: To reduce the influence of the semantic gap in image retrieval, this paper presents an automatic image annotation method combining aspects and visual semantics.This method captures the latent aspects from the textual space of the training image set using probabilistic latent semantic analysis model firstly.And then, Gaussian Mixture Model of the each latent aspect is constructed according to the high dimensional image visual feature, describing the visual semantic content of each aspect.This method reduces the semantic gap, and improves the accuracy of the automatic image annotation.This method is compared with several other state-of-the-art methods on the standard Corel dataset.The results of experiments show that this method achieves better average recall and better average precision.The effectiveness of this method has been proved.