A tensor decomposition based clustering method was proposed for heterogeneous information in networks. This clustering method can cluster multiple types of objects and rich semantic relationships simultaneously. The multi-types of information objects in networks were modeled as a high-dimensional tensor, and the rich semantic relationships among different types of objects were modeled as elements in the tensor. Based on an effective tensor decomposition method, the multi-types of objects were partitioned into different clusters simultaneously. The experimental results on both synthetic datasets and real-world dataset show that the proposed clustering method can deal with the heterogeneous information in networks well, and can outperform the state-of-the-art clustering algorithms.