An adaptive sparse representation regularized reconstruction method for accurate parallel imaging was proposed by exploring the strength of dictionary learning in capturing image fine structures while promoting sparsity. The reconstruction was formulated as a minimization problem, which consisted of a data-fidelity term and a dictionary learning term and was solved by the “divide and conquer” strategy. The comparative results of the proposed method with respect to two popular approaches on an in-vivo dataset demonstrated that the proposed method preserves more image fine details while suppressing noise.