In phase shifting interferometry, surface measurement error will exist due to the phase shifting error resulted from the system or environment. The principle of the least-squares iterative algorithm with randomly phase shifting is introduced. The original surface can be reconstructed with high accuracy by iteration in the numerical simulation. The phenomenon that the reconstructed surface may be inversed to the actual one because of the randomly phase shifting is analyzed. It is proposed that the phase shifting orientation should be known in this algorithm to achieve correct results. Experiments are carried out for verification. The calibration method of the phase shifting direction of the interferometer is indicated.