Abstract: The effect of laser power,scanning speed and powder feeding rate on the depositing height of single pass welding is firstly analyzed in laser direct fabrication(LDF),and the result shows that there is the non-linear intrinsic relationship between them.Based on the non-linear characteristic,the least square support vector machine(LS-SVM) network is adopted to predict the building precision.Training and testing samples are collected by using single pass experiments,and non-linear mapping relationship between them is built.Then,the trained LS-SVM network is examined by testing samples,and their training and testing performances are investigated.The result shows that generalization ability,function-approximating ability and real-time of the LS-SVM network are better for the prediction of the building precision in LDF,and high definition thin-walled parts are successfully fabricated.