Energy-based multisource localization is an important research problem in wireless sensor networks (WSNs). Existing algorithms for this problem, such as multiresolution (MR) search and exhaustive search methods, are of either high computational complexity or low estimation accuracy. In this paper, an efficient expectation-maximization (EM) algorithm for maximum-likelihood (ML) estimation is presented for energy-based multisource localization in WSNs using acoustic sensors. The basic idea of the algorithm is to decompose each sensor's energy measurement, which is a superimposition of energy signals emitted from multiple sources, into components, each of which corresponds to an individual source, and then estimate the source parameters, such as source energy and location, as well as the decay factor of the signal during propagation. An efficient sequential dominant-source (SDS) initialization scheme and an incremental parameterized search refinement scheme are introduced to speed up the algorithm and improve the estimation accuracy. Theoretic analyses on the algorithm convergence rate, the Cramer–Rao lower bound (CRLB) for localization accuracy, and the computational complexity of the algorithm are also given. The simulation results show that the proposed EM algorithm provides a good tradeoff between estimation accuracy and computational complexity.