This paper proposes a hierarchical clustering algorithm based on information granularity, which regards clustering on sample data as the procedure of granule merging. In the promoted algorithm, firstly each sample is named with an initial class, then for a given granular threshold those pairs of samples, whose distance among them is less than the threshold, will be merged to one class and generate a new larger granule. Repeat this procedure until certain conditions are satisfied. This paper also discusses com- putational complexity of the novel algorithm and compares them with the traditional hierarchical clustering algorithm. In the last, some experimental examples are given, and the experimental results show that this algorithm can efficiently improve the clustering speed without affecting the precision.