A low-complexity algorithm for Directionof arrival (DOA) estimation based on Conjugate gradient(CG) method is proposed in this paper. The orthogonalresidual vectors in CG method span the signal subspaceemployed by the Multiple signal classification (MUISC)spectrum. The computational complexity of the proposedalgorithm is greatly reduced, since it does not involve theestimation of the covariance matrix of observation dataand Eigenvalue decomposition (EVD). Besides, a new criterion for the dimension estimation of the signal subspaceis also proposed, which can significantly overcome the leakbetween the signal subspace and noise subspace in theabsence of the knowledge of source signal enumerationand desired signals. Simulation results are presented todemonstrate the approximate performance of the proposedmethod comparable to the traditional EVD method, andthe significant reduction in computational complexity dueto the use of finite conjugate iterations.