A method for improving the measuring accuracy of structured light measurement system, which adopts projecting stripe pattern to measure the three-dimensional profile, is presented. Based on the evaluation of the reliability of center extraction results, the improvement of accuracy is achieved by identifying and rejecting the stripe center extraction results with large error. Two parameters are used to evaluate the reliability of center extraction results. The first parameter is the average energy of the stripe, which is used to analyze and establish the relationship between the extraction accuracy and the signal-to-noise ratio through a statistical method. The second parameter is the asymmetric degree of the stripe gray distribution which introduces error into the center extraction, and a new method is proposed for measuring the asymmetric degree. Then, the criteria of the data rejection defined by the thresholds are presented, and large error data with low reliability are identified according to the thresholds. Higher measuring accuracy is achieved by rejecting the identified data. The validity of the method has been proved by experiments.