Community detection in complex networks is a challenging task. In this work we introduce a novel latent communities model (LCM) that models the communities as latent factors of networks and reveals the communities by expectation-maximization algorithm. LCM can model the community membership explicitly with values between zero and one. What's more, we propose a modified purity to evaluate the performance of community detection. Experimental results on both synthetic network and real-world networks demonstrate that LCM can achieve comparable performance.