The function optimization problem for analyzing the data of river stream water quality test to estimate such water quality parameters as longitudinal dispersion coefficient, average stream velocity and the other parameter is to be solved with simulated annealing method. Such measures as adding supplement constraint condition on state variables, setting up interior valve value and adding memory function in interior loops are taken to improve convergence of algorithm. The example shows that these improvement measures are very effective and the improved algorithm may give better estimation results of water quality parameters. On the other hand, the example also shows that the length of Markov chain (the times of interior loop) bring litter effect on the times of outer loops in the investigated problem. Finally, because that there is no supplement restriction to the objective function, and the initial values of parameters to be estimated being generated in random means, the improved SA algorithm will be widely applied to estimation of water quality parameters in analyzing the data of river water quality tests along with further and more research work to be done on this subject.