提出了构造性RBF神经网络的一种新的全监督式学习算法 .该算法在神经网络隐层引入新节点 ,并通过使新节点的输出方向尽力逼近学习残差的方向获取网络参数 ,从而减少学习误差 ,实现对训练样本的学习 .仿真结果表明了该新学习算法的有效性 . This paper proposes a new all-supervised learning algorithm of structured RBF neural network which introduces a new node in the hidden layer of neural network and obtains the network parameters by making the output direction of the new node try to approach the direction of learning residual to reduce Learning error and learning of training samples.The simulation results show the effectiveness of the new learning algorithm.