Abstract: Via near infrared(NIR) spectroscopy combined with radial basis function neural network(RBFNN), a model for determining pyrazinamide(PZA) content in tablets was established. Leave-one-cross-validation method was used for selecting the most effective preprocessing method, the most suitable topological parameters and the best spread constant in the RBFNN. The results showed that Savitzky-Golay smoothing method was the most effective preprocessing method, and the most suitable number of input nodes and hidden nodes were 8 and 15 respectively, and the best spread constant was 2.5. The optimum parameters were applied to establish the model for determining the pyrazinamide content in tablets with the root mean squares error cross-validation(RMSECV) of 0.005 52. Using this model for determining the pyrazinamide content in the prediction set, the root mean squares error of prediction set(RMSEP) was 0.003 30 and the average recovery was 100.091%. NIR method and UV-spectroscopy method were used for determining the pyrazinamide content in 6 different batches of Pyrazinamide tablets respectively, and the relative error between the values obtained by these methods was less than 4.179%. These results demonstrate that the NIR method is precise, convenient, and rapid, and involves no pretreatment and pollution, and may have extensive application in pharmaceutics quantitative analysis.