Abstract: Genetic algorithm was used to optimize the backpropagation neural network connection weights and improve the models predicted precision and generalization ability on the basic of the mechanical properties prediction models of welded joints established with back-propagation neural network.The performance analysis shows that the predicted trend agrees well with the previous research work and the predicted error is less than 5%.It is obvious that the models will be more applicable and valuable in the practice with the enlargement of database and the data-covering space