Laser beam pointing systems are generally affected by pointing errors. The pointing errors cause energy loss and the decline of the system performance while the center of the laser beam is not on the target. The pointing errors can be estimated from the scatter return photons back from the illuminated target. The performance of pointing error can be improved with close-loop control on the measurement of the return signals. In this paper a method to reduce pointing error based on the return signals from the illuminated target is studied. The stochastic parallel gradient descent (SPGD) algorithm is applied in this method. Some effects to influence the precision of pointing errors are analyzed. The analysis is verified by experiment.