The feasibility of predicting the octane number by the near infrared spectra analysis technique in the quantitative analysis was investigated. 12 calibration gasoline samples and 4 prediction samples were collected and their absorption spectra were measured with a high precision model Lambda 19 scanning spectrophotometer. Based on the multivariate statistic model, stagewise regression and partial least squares were selected. A model of the correlation behavior between the spectra and octane number was established and calibrated, and then the prediction of octane number is made. The prediction error is +/-1.0 octane number in the best analysis results.