In this paper, the closely spaced coherentsource localization is considered, and an improved method based on the group delay of Multiple Signal Classification (MUSIC) is presented. Firstly, we introduce the spatial smoothing technique into direction of arrival (DoA) estimation to get rid of the coherent part of signals. Due to the degraded performance of subspace based methods on the condition of nearby sources, we then utilize the MUSICGroup Delay algorithm to distinguish the closely spaced sources, which can resolve spatially close sources by the use of the group delay function computed from the MUSIC phase spectrum for efficient DoA estimation owing to its spatial additive property. Theoretical analysis and simulation results demonstrate that the proposed approach can estimate the DoA of the coherent close signal sources more precisely and have higher resolution compared with subspace based methods.