Networks are widely used in a variety of different fields and attract more and more researchers. Community detection, one of the research hotspots, can identify salient structure and relations among individuals from the networks. Many different solutions have been put forward to detect communities. EM as a model on statistical inference methods has received more attention because of its simple and efficient structure. Unlike many other statistical inference methods, no extra information is assumed except for the network itself and the number of groups for the EM approach. However, practical usefulness of the EM method is often limited by computational inefficiency. The EM method makes a pass through all of the available data in every iteration. Thus, if the size of the networks is large, every iteration can be computationally intensive. Therefore we put forward an incremental EM method-IEM for community detection. IEM uses the machinery of probabilistic mixture models and the incremental EM algorithm to generalize a feasible model fit the observed network without prior knowledge except the networks and the number of groups. Using only a subset rather than the entire networks allows for significant computational improvements since many fewer data points need to be evaluated in every iteration. We also argue that one can select the subsets intelligently by appealing to EM’s highly-appreciated likelihood judgment condition and increment factor. We perform some experimental studies, on several datasets, to demonstrate that our IEM can detect communities correctly and prove to be efficient.