Abstract: A new two-step medical image retrieval algorithm for clinical practice was proposed combining the gray level co-occurrence matrix with gradient phase mutual information. Based on the good classification performance of gray level co-occurrence matrix, the algorithm refines the retrieved process to improve its precision and the integral robustness of the retrieval algorithm. With the algorithm applied to 6 different anatomical positions of CT images, the testing results showed that the algorithm can provide high precision while attaining near real-time speed of response to queries. Expanding properly the algorithm, it can be easily applied to clinical practice in wider fields.