The pollution concentration prediction model of lubricants based on the principle of reflection spectrum is studied. Pure oil and full pollution oil are chosen as experimental materials. Different concentration samples are gradually prepared by equivalent volume method. The spectrometer of 200~850 nm band range is used to measure the samples. Xenon lamp serves as excitation light source. After adjusting, remain unchanged the distance and geometry azimuth between the detector probe and the sample, spectral measurement can be just conducted. The spectral reflectance experimental results show the higher contamination of the lubricants, the lower intensity of the reflected spectrum. In the variable step size algorithm on the original experimental data sparse sampling based, Correlation Coefficient, PCA, and PCA combined with Correlation Coefficient method (PCA-Correlation Coefficient) are respectively adopted to select advantage characteristic wavelength of lubricating oil in the wide band of 220~780 nm, and the index regression model is established about oil pollution concentration and spectral reflectance. Test results show that when the concentration is greater than 006, the index regression model of pollution concentration and spectral reflectance at 37893 nm which is selected by PCA-Correlation Coefficient can well realize the prediction and estimation of pollution concentration of lubricant oil. So the regression model based on PCA-Correlation Coefficient to select advantage wavelength is suitable for the quality estimation of lubricating oil with high pollution concentration. Because of the specific conditions of the concentration and mixed medium, the model does not obey Lambert-Beer Law, which apply to low concentration, uniform and transparent solution. The paper provides a feasible experimental basis for the further to achieve the working lubricants pollution concentration with reflection spectrum method online, fast and accurate determination.