A calibration model (NIRS-RBFNN) based on combination of near infrared reflectance spectroscopy and radial basis function neural network (RBFNN) has been proposed for synchronous and rapid non-destructive determination of Coriolus versicolor. Savitzky-Golay smoothing (SGS), fast Fourier transform (FFT), derivative of wavelet transformation (WT) and wavelet packet transformation (WPT) with multi-scale analysis were used to dispose the original NIRS, then principal component analysis (PCA) method is used to obtain the principal components (PC) scores. The anterior 15 PC scores were used as input data. These developed RBFNN have been optimized by selecting suitable parameters of input data, numbers of hidden layer neurons and spread constant through different pretreated spectra of calibration. The optimal quantitative analysis (NIRS-RBFNN) model for polysaccharide and protein of coriolus versicolor: for polysaccharide the optimal model is 6 scales reconstructed spectra of WPT, model parameter is WPT-NIRS-RBFNN(7-12-1, 3.2)，and root mean squared error of cross validation(RMSECV) is 0.009897, Rcv=0.98357; root mean squared error of predictions(RMSEP) is 0.00909; Rp=0.98283. For Protein the optimal model is 6 scales reconstructed spectra of WPT, model parameter is WPT-NIRS-RBFNN(12-10-1, 3.0)，and RMSECV is 0.00524, Rcv=0.99426; RMSEP is 0.00998; Rp=0.98246. These results show that the model has good robustness and precision and NIRS technology is convenient, rapid, no pretreatment and no pollution that this method could be popularized in the in situ measurement and the on-line quality control for coriolus versicolor.