We have researched subjective interpretation for statistical texture images in order to clarify human subjective interpretation mechanism for images. We have focused on orientation and color features, and unified those features by calculating the contrast over the entire image resolution. Next, we have measured psychological responses and corresponded them to our texture feature based on the canonical correlation statistics. We developed the STIR (subjective texture image retrieval) system based on our model. The results show that STIR can retrieve images which give a similar impression. Moreover, STIR can retrieve images adaptively to user's preference by utilizing user's subjective model.