The diesel fuel properties are closely related to the performance of diesel engine. It is of great significance to quickly and accurately detect the multi-properties of diesel oil. The purpose of this paper was to detect 50% distillation temperature, cetane number, viscosity and freezing temperature of diesel oil by using chemometrics methods combined with near infrared (NIR) spectroscopy. Savitzky-golay (SG) smoothing and first derivative (1d) were used to denoise the spectrum, and Monte Carlo sampling method was applied to detect and eliminate outliers. Two wavelength selection algorithms, stability competitive adaptive reweighted sampling (SCARS) and correlation coefficient (CC), were proposed to extract feature wavelength points of diesel oil. Then SCARS-SVM and CC-SVM prediction models were established, respectively. In the prediction of four properties of diesel oil, SCARS-SVM has lower root mean square error of prediction (RMSE) and higher coefficient of determination (R2) than CC-SVM, especially for mean absolute percentage deviation (MAPD), SCARS-SVM is several times or even dozens of times lower than CC-SVM. In addition, the running time of SCARS-SVM is shorter than that of CC-SVM. The satisfying results demonstrated that SCARS-SVM model is more suitable for quantitative analysis of diesel multi-parameter properties.