Weighted semblance can be used for improving the performance of traditional semblance for specific data sets. We have developed a novel approach for prestack velocity analysis using weighted semblance. The novelty came from a different weighting criteria in which the local similarity between each trace and a reference trace is used. On the one hand, low similarity corresponded to a noise point or a point indicating incorrect moveout, which should be given a small weight. On the other hand, high similarity corresponded to a point indicating correct moveout that should be given a high weight. Our approach could also be effectively used for analyzing AVO anomalies with increased resolution compared with AB semblance. Synthetic and field common-midpoint gathers demonstrated higher resolution using the approach we developed. Applications of the proposed method on a prestack data set further confirmed that the stacked data using the similarity-weighted semblance could obtain better energy-focused events, which indicated a more precise velocity picking.