An efficient method was presented to select wavelength regions of NIR (near infrared spectroscopy) based on WT(wavelet transform) for building a PLS(partial least squares) calibration model. Wavelet approximation coefficient (WAC) is similar matrix of original NIR. By linking the correlation coefficients of the optimal WAC with original NIR data, a research space with the all combinations of strong correlativity wavelength was selected as final wavelength regions to build a PLS calibration model of NIR. This selecting method considered the influence of density matrix on wavelength selection, filtered wavelet detail coefficient entirely, and avoided the interference of high frequency noise. The running time of building model was reduced enormously so that the final model is of higher accuracy. Compared with other traditional wavelength selecting method, This method is validated in the measurement of rice apparent amylose content by NIR. The apparent amylase content test results showed that the number of wavelengths for building the models can be reduced to 20% of the original method, the calibration model and the prediction precision are greatly improved by wavelet transform algorithm.