Pavement void is one of the important parameters of influencing airport operation safety. In order to study the non-destructive test, back propagation prediction method research of pavement void on the basis of wavelet transform was proposed. By indoor model test, impacting load was imposed on scaled pavement model and vertical acceleration signals of pavement were obtained, which were used to achieve power spectrum analysis, energy spectrum analysis and time-scale analysis by wavelet transform method. Back propagation(BP) neural network  was trained by 1500 sets of feature vectors of indicating different pavement void conditions in order to improve the prediction function. Site experiment of airport pavement was conducted and pavement concrete samples were drilled to validate the reliability of BP neural network prediction. The results showed that, pavement vibration signals reflected the dissipative effects of energy signal in void area after wavelet transform analysis. It had significant differences and regular patterns in completed void and half void areas. The results of site experiment were satisfied the prediction of BP neural network trained by indoor test results. And signal of slight void was distinguished, by which the reliability and feasibility were proved in pavement void evaluation by this method.