Attenuated total reflection-Fourier mid-infrared spectroscopy（ATR-FTIR） and competitive adaptive weighted sampling（CARS）， combined with physical and chemical analysis， were established to quickly determine the content of water soluble protein and total sugar in royal jelly. Based on the CARS method， Savitzky-Golay smoothing and first-order derivatives were used to preprocess the original spectrum， and 84 characteristic wavelength points were selected. The model has the best prediction effect on water soluble proteins. The correlation coefficients of the correction set and the verification set are 0.9631 and 0.9771， respectively， RMSECV and RMSEP are 0.0913 and 0.0754， respectively. The PLS quantitative model of total sugar screened 111 characteristic wavelength points and subjected to Savitzky-Golay smoothing and second-order derivative processing. The results were the best. The correlation coefficients of the correction set and the verification set were 0.9750 and 0.9850， and the RMSECV and RMSEP were only 0.4485 and 0.4244， the modeling effect was good. It can be seen that based on the CARS method for screening characteristic wavelengths， it is feasible to use ATR-FTIR and partial least squares to quickly detect the content of water-soluble protein and total sugar in royal jelly， and the established quantitative model has good robustness and high prediction accuracy.