Abstract: A pile detection method combining wavelet analysis and neural network is introduced. According to the characteristics of ultrasonic propagation in the pile foundation, the collected ultrasonic signals are analyzed by using wavelet analysis. The method performs wavelet packet decomposition, normalizes the decomposed signal and constructs a feature vector of ultrasonic signal to characterize pile foundation defect information. Furthermore, multiple sets of feature vectors were taken as training samples of the neural network in order to train and learn the feature vectors. The non-diagnosed samples were input into the neural network for identification verification. Experimental data shows that the trained neural network can effectively identify pile foundation defects and defect types.