Abstract: The traditional image retrieval needs to compute the distance between each picture in the database with the requested picture. The retrieval speed and accuracy is bad. Aiming at this drawback, this paper presented a novel image retrieval method based on improved GHSOM clustering algorithm. The feature of images formed clusters by improved GHSOM algorithm. The first step of the image retrieval is to find out the cluster which is similar to the request picture in the GHSOM network. Then continue to retrieval the images in this cluster. The retrieval speed improved greatly. In the process of searching for similar cluster, the hierarchical structure of GHSOM network was made full use of to improve the retrieval efficiency. AIC criterion was created according to the characteristics of the algorithm. The improved GHSOM algorithm applied this criterion to select proper growth parameter for each SOM map. Proper growth parameters make each map can represent data set well. The accuracy of image retrieve will improve too. Experiments illustrate that the improved GHSOM algorithm can gain a better cluster performance. The speed of image matching has improved nearly three times. Image retrieval accuracy has also been improved to some extent.