Abstract: On purpose to apply rational function for mathematical model building,this article proposes a practical relaxation algorithm to optimize both parameters and structure of rational model at the same time.First,structural pattern recognition techniques toanalyse objective curve and find a set of important sample points of the curve isintroduced,then primary structure is driven for optimization of structure from thepoint set.The primary structure also can offer good initial approximation functionfor parameter optimization.For parameter optimization,in this article a practicalleast square estimation formula based on Hamming method is given.It is shown bypractical applications that this algorithm has the advantage of high speed,high precision,fewer number of produced model parameters and strong adaptiveness to objective problems.