The main problem in content–based medical image retrieval system is semantic gap, where the meaning that the user has in mind for an image is at a higher semantic level than the features on which the database operates. To overcome the shortcomings in the current content–based medical image retrieval systems and to provide a mechanism to better understand the images semantics, a fuzzy rule based method is proposed which determines which of the image features are more important than the other ones, by making a proper weight vector for the distance measure. For instance, for a given query image, large weights could be assigned to shape features, whilst texture features could be almost ignored by taking small weights. For the training purpose, an algorithm is provided by which the system adjusts its fuzzy rule parameters by gathering the trainers opinions on which and how much the image pairs are relevant. For further improving the performance of the system, a feature space dimensionality reduction method is also proposed. To ensure that this method will increase the precision of the system, we monitored the precision parameter in its training. Our experimentation on IRMA medical image data set shows that the proposed method outperforms two of the most popular image retrieval methods, i.e. a classification based method and a feature weighting technique, and could be used to reduce the semantic gap in the image retrieval systems.