The detection of oil content in soil has an important practical significance in oil pollution prevention and control. We measure the hyperspectral reflectivity and the oil content for soil samples in Gudong Oilfield. Using variable forecast model and stepwise regression method

we analyze the linear and nonlinear relationships between soil spectral characteristic parameters and oil content. The experiment shows that there is a significant correlation between the third broken line segment slope of envelope line analysis and the oil content. The cubic function of this section slope is the best single variate estimation model. The standard normal variate transformation has the best effect on spectrum pretreatment. When the transformed spectral are used to build multivariate regression model

the adjusted coefficient of determination is 0.826

and the total RMSE is 0.531

which is the best forecast model. The method of using hyperspectral data to detect the oil content will provide an effective new way for detecting the oil pollution in soil.