Based on different characteristics analysis of the systematic and random errors found in the spectral measurement data, two methods are put forward to diminish their impacts on the reverse determination accuracy of thin films’ respectively optical parameters. The first method uses spectral coefficient’s first order partial derivatives of thin films with respect to layer optical parameters to select measurement data. The spectral region or incident angle region characterized with opposite operation symbol of partial derivatives is chosen as the optimal measurement data region to minimize the deviations from the real optical parameters of thin films caused by systematic measurement errors. The second method injects independent random noise of same distribution to the actual measurement data. The data with noise are utilized as the real spectral coefficients of the thin film being measured in the reverse determination for many times. The reverse determination algorithm used is a mixed optimization method based on the discrepancy function minimization technique. The statistics mean values of all fit optical parameters obtained each time by the mixed optimization method are chosen as estimates of the real optical parameters of the measured thin film to diminish or eliminate the uncertainty of optical parameters caused by random measurement errors. These two methods are of obviously practical value to improve the determination accuracy of thins films’ optical parameters. And they also characterize with concise and definite physical meaning, expedient manipulation ability and good universality, not restricted to specific spectral regions or specific materials. They are of promising importance in the accurate reverse determination techniques of thins films’ optical parameters.