Abstract: In order to improve recovery accuracy for compressed sensing, a Stagewise Weak selection Modifying approximation Conjugate Gradient Pursuit (StWMCGP) algorithm is proposed in this paper. This algorithm modifies the direction in the directional pursuit algorithm and clearly presents a stopping criterion to search the indices of elements and get a set. Then the evaluation of sparse signal is obtained by using Least-squares algorithm and the set. Simulated results show that for the same sparsity level, the number of measurements needed by the algorithm is about 20% less than that needed by MP or StOMP-FDR to exactly recover. When recovering two-dimensional image signal, the recovery accuracy of this algorithm is about 1% higher than that of BCS or StOMP-FAR.