Phase retrieval algorithm is widely used to eliminate the unwanted conjugate image for in-line digital holography. Compared with the phase retrieval algorithms based on single hologram, phase retrieval with images recorded at multiple distances could improve the precision and convergence speed apparently, especially for the reconstruction of two images. However, it is critical to optimize the recording distances for the two images to achieve high precision and the true value of the reconstructed images. We record a lot of digital holograms at axially different planes in experiment. After analyzing the normalized root-mean-square (NRMS) error of the data, we conclude that the algorithm is suitable for the distance between 130 mm and 160 mm. And the conclusion can also be made that the reconstructed image which has the highest precision is achieved when the object distance is 150 mm and the object distance interval is 2 mm for which the NRMS error is 0.0096.