In this paper, we combine the use of reduced feature vector integration (RFI) and distance integration (DI) with relevance feedback (RF) on 3D model similarity retrieval. The RFI outperforms the individual FVs and gives high probability of providing relevant objects, other than the query itself, on the limited-size display window. Therefore, user may select the relevant object(s) just after the initial query. The DI enhances the precision by estimating the weighting factor from the variance of the distance and the rank of relevant objects, and pushing the relevant objects to the top and the irrelevant objects to the bottom. By utilizing both approaches, the smal number of RF iterations significantly improves the retrieval precision.