Cluster synchronization plays an important role in the proper functioning of many real-world network systems. The mechanism behind the cluster’s emergence and transformation in response to parameter change, especially in heterogeneous networks that lack symmetry, is crucial to the general understanding of collective behavior in realistic large network systems and yet has remained out of reach by existing approaches. We uncover a mechanism for cluster synchronization in heterogeneous networks by developing a heterogeneous mean field approach along with a self-consistent theory to analyze the cluster structure and its stability.