To detect rapidly the oil yield of an oil shale on a single point measurement, a detection method was proposed based on combination of wavelengths by a portable near infrared spectroscopy. By taking the kaolin mixed with oil simulating samples as the research object, how to select the combination of wavelengths by near infrared spectroscopy was performed. With proposed method, the full spectral reflectivity data were firstly acquired by the developed portable spectrometer. Then the preliminary selection for the combination of wavelengths was implemented by differential multiplicative scatter correction method and correlation coefficient method, and the optimum combination of wavelengths was determined by combination generation method and leave one cross validation Multiple Linear Regression (MLR). Finally, the data for the optimum combination of wavelengths were verified by MLR modeling. The results show that the optimum combination of wavelengths is 1 644, 1 720, 2 210 and 2 260 nm, and the determination coefficients for calibration set (R2c), prediction set (R2p ) of 30 modeling samples and validation (R2p) of 11 validation samples are 0.995 4, 0.997 7 and 0.990 1, respectively. The method provides a basis for design of a rapid detecting spectrometer in oil yields of oil shapes.