The paper uses digital image processing technology, technology of face pattern recognition and traditional database retrieval technology, integrate image retrieval technology based on version and content, and avoid the complexity of matching image process. The experiments with 200 human samples, correctly retrieved for 155 people, exactly matched to 125. The recognition rate of the system is 75.55%, the average time of search is less than 0.1s. Experiments indicate this method has strong robustness. The semantic face image retrieval system using this method has the characteristics of fast, efficient, practical.