The evaluation function of multi-layer optical thin film systems is a complex nonlinear multi-modal function. Traditional numerical methods could hardly obtain global extreme when the properties of evaluation function are unknown, with calculation of instability and low efficiency. Simulated annealing algorithm (SAA) exhibites some advantages such as high efficiency, less restricted by the initial solution and ability of obtaining global extreme. This is because SAA acceptes the optimized solution and part of the poor solution at the same time. General simulated annealing algorithm is improved by adding heating and memory module. Improved algorithm could select the initial value of control parameters more reasonably and memorize the optimal solution ever encountered in search process. Thus the ability to jump out of local optimum “trap” is improved accordingly. Antireflective film, neutral spectroscope and 10.4 μm long-wave pass filter desiged by improved algorithm, the value of evaluation function is decreased by 77.6%, 31.9%, 96.2% respectively. Film system has good spectral characteristic and more flat transmittance curves.