In consideration of the serious size distribution broadening and tailing phenomena come from improper iterative steps in inversing laser particle size distribution, this paper presents a method to select the iterative step for Projection algorithms to improve measurement accuracy. Based on the similarity theory, this method obtains the similarity between the row vector of light energy distribution and the course quantity of light energy coefficient matrix. Then, it gets the predict distribution of particle size after filtering and normalized processing. Furthermore, according to the predicted result, the iteration step can be calculated. The measurement results for national standard particles GWB(E)120046 show that the measuring errors for D50, D10, and D90 are -0.6%, -1.1% and -0.6%, respectively, and the light energy logarithm error of mix standard particles GWB(E)120041 and GWB(E)120049 is 2.167. Results demonstrate that the algorithm can effectively improve noise immunity, measurement accuracy and resolution.