This paper presents an efficient correspondence grouping algorithm to search inliers from an initial set of feature matches. The novelty lies in the proposal of a scoring technique for measuring the reliability of a triple combination (three pairs of matches) based on redundant geometric constraints. According to the proposed scoring method, several top-ranking triple combinations are selected for estimating the transformation hypotheses between two 3D shapes. For each transformation hypothesis, a correspondence from a selected correspondence set should cast a vote whether it is satisfying the geometric constraint with it. Finally, the transformation hypothesis with the most votes is considered as the best transformation and the correspondences from the initial correspondence set agreeing with the best transformation are grouped as inliers. We performed both comparative experiments and real application experiments to evaluate the performance of our proposed method on five popular datasets. The experimental results show the superior performance of our method with respect to different levels of noise, point density variation, partial overlap, clutter and occlusion. In addition, our proposed method can boost the performance of a feature-based 3D object recognition algorithm, giving an increase in both high recognition rate and computational efficiency.