针对超声无损检测中缺陷分类难、分类结果可靠性差等问题，给出了一种以神经网络为基础的缺陷特征分类方法．利用Ｆｉｓｈｅｒ线性判别方法对表征缺陷特征的时域信号的波形参数进行了选择，并将这些参数作为神经网络的输入矢量对网络进行训练．用该网络对缺陷特征进行了识别，结果表明：神经网络的识别率远大于传统的贝叶斯分类方法． Aiming at the defect classification of ultrasonic non-destructive testing and the poor reliability of classification results, a method of defect feature classification based on neural network is proposed. The Fisher linear discriminant method is used to select the waveform parameters of the time-domain signal characterizing the defect features, and these parameters are used as the input vector of the neural network to train the network. The characteristics of the defect were identified by the network. The results show that the recognition rate of neural network is far greater than the traditional Bayesian classification method.