Simulated annealing algorithm (SAA) was applied to locate the global minimum of a non-linear least square function for spectra fitting. The resolution of methylene region of two13C-NMR spectra of polystyrene was implemented successfully with this method. Compared with the existing methods, this technique has important advantages in two respects: one is that the resulting resolution of spectra with this method have lesser dependence on the initial values of the parameters inputted to start the optimization of the least square function; the other is that it provides the possibility to determine the peaks which cannot be recognized by the observation. With this technique, some possible error arising from the observation can be avoided.