Abstract: Global market competition among researchers and manufacturers has prompted the need for developing new material at reduced cost and rapid response to customers′ demands, but the conventional methods which depend on experience and lots of experiments do not meet these situations. In this paper, a prediction model based on SLFM network is proposed to predict material properties using only a few of experiments. The topology, algorithms of learning and prediction, and parameter selection of this network are discussed when the prediction model is implemented. An engineering application example on Ti 26 alloy is given to show that this prediction model possesses advantages of high speed of learning (74 iterations for 20 learning samples), high learning accuracy(100 percent in recalling accuracy), and quite good prediction performance(the value of relative error can be mostly controlled in 3%). Based on its good performance, part factors affecting the properties of Ti 26 alloy are optimized. The prediction results and practical results show good agreement. This prediction model has been proved to be an available tool to solve the problems on prediction and optimization for new material development. It also supplies a new idea for prediction and optimization of multivariate and nonlinear systems.