This paper presents a new iterative algorithm called constraint removal (CR) for the recovery of a sparse signal x from an incomplete number of linear measurements y such that ym× 1= Am× nxn× 1 and m<n. It is empirically demonstrated that the CR algorithm has a recovery performance which is between basis pursuit linear programming (BP-LP) and subspace pursuit (SP) for both zero-one and Gaussian type signals.