In this work, the near infrared spectroscopy (NIR) method to determine the diesel group composition has been successfully designed. The 93 representative sample of diesel oil are selected to determine the group composition. These samples are covering a wide range and the concentration is evenly distributed. The multiple linear regression method is chosen to establish the calibration model. 17 samples are used to verify the calibration model. The paired t test statistic is 0.05. It is smaller the t-value 2.12, when the significance level is 0.05. The repeatability of the NIR method is better than the standard method. Compared to the reference method, the NIR method is simple, fast; do not consume the sample and organic solvents. This NIR method to determine the diesel group composition is feasible and repeatable and can meet the requirement in the small oil refinery.