In this paper

a new greedy refinement bayesian approach (GRBA)

used to solve the joint sparse signal recovery problem

is proposed.The joint sparse property of signals is first used to model the signals.Based on the model

a greedy Bayesian inference method used to estimate the signals is then presented.In order to enhance the performance of the recovery

the covariance matrix got by the Bayesian inference is utilized to refine the support recovery results in our inference process.The analytical results show that GRBA outperforms the reported algorithms in the literature in terms of both the signal recovery accuracy and computational complexity.It keeps both the advantages of Bayesian methods and greedy methods.Numerical simulations verify the effectiveness of the analytical results.