The work aims to propose an image retrieval algorithm based on dictionary statistics and normalized multiple distance with respect to the disturbance easily caused to the current color image retrieval technology, thus leading to its inadequate robustness and other deficiencies. Firstly, the image was quantized and converted into a one-dimensional signal. Then the dictionary statistics were introduced to code the one-dimensional signals. The diversity value of the image were calculated. At last, on the basis of normalized dictionary distance, the diversity values of image obtained through dictionary coding were inserted and thus the similarity metric criterion of normalized multiple distance (NMD) was defined. NMD was adopted to compare and identify the diversity values of the query image and database image, and search out the image most similar to the query image with the same characteristics, so as to complete the retrieval task. Experimental results from the COREL database showed that: relative to the current commonly used retrieval techniques, the proposed retrieval algorithm had better precision ratio and recall ratio and could accurately retrieve color images, which effectively reduced the disturbance caused to retrieval performance by the color. In conclusion, the proposed algorithm has better retrieval precision and can be used for retrieval tasks in the fields of medical treatment and trademark, etc. in a better manner.