Stoilov algorithm is a recently developed equivalent step algorithm whose phase step is arbitrary. The value of the phase shifting needn’t be known but phase step must remain invariable. The wrap phase caused by measured object can be derived correctly. So it is paid wide attention in three-dimensional measurement field. However, Stoilov algorithm excessively depends on the intensity of the image, and subtraction, division and extraction must be operated in its mathematical model. Imaging process will cause some abnormal phenomena somewhere such as the zeroth denominator and complex phase, which lead to the incorrect phase calculation or incorrect phase unwrapping. The reconstructed object will be misshapen or anamorphic, or even the measured object can’t be reconstructed. So a new method based on the statistical approach to improve Stoilov algorithm has been proposed. By this method the phase error brought by the abnormal phenomena is restrained and the precision of three-dimensional measurement is improved. Experiments show its feasibility and validity.