Clustering is a fundamental analysis tool aiming at classifying data points into groups based on their similarity or distance. It has found successful applications in all natural and social sciences, including biology, physics, economics, chemistry, astronomy, psychology, and so on. Among numerous existent algorithms, hierarchical clustering algorithms are of a particular advantage as they can provide results under different resolutions without any predetermined number of clusters and unfold the organization of resulted clusters. At the same time, they suffer a variety of drawbacks and thus are either time-consuming or inaccurate. We propose a novel hierarchical clustering approach on the basis of a simple hypothesis that two reciprocal nearest data points should be grouped in one cluster. Extensive tests on data sets across multiple domains show that our method is much faster and more accurate than the state-of-the-art benchmarks. We further extend our method to deal with the community detection problem in real networks, achieving remarkably better results in comparison with the well-known Girvan-Newman algorithm.