Two frequent problems in content-based image retrieval systems are the poor quality of the results, due to the semantic gap, and the slowness of the image retrieval. For controlling these problems when human face image sets are considered, a new approach, called the MIFLIR system, is presented in this paper. This approach allows reducing the semantic gap by using fuzzy predicates during the indexing and the retrieval of images. In addition to this, our system innovates by using a metric indexing method to accelerate perceptibly the retrieval of images. Experimental results show that the MIFLIR system reduces noticeably the total time for performing fuzzy queries, when compared to a similar system which uses sequential scan search method.