Rapidly and accurately detection of the type and content of mineral oil in water pollution has important significance for the timely screening and control of pollution sources. The use of infrared spectral analysisi technology to detect mineral oil has advantanges of efficient, fast and pollution-free. Infrared spectrum technology is very for the detection of mineral oil in the water. In order to obtain a more reliable results, Fourier transforms attenuated total reflection infrared spectrometry (FITR-ATR) technology is used to get the spectral information of the mineral oil sample, and SPXY method is used to divide the sample set. The paper not only analyzed partial least squares (PLS) and iterative Bagging partial least squares (IBPLS) the two different methods to build regression model, also compared the difference of using the method of the combination of Savitzky-Golay (SG) smoothing and the method of a single iterative Bagging partial least squares (IBPLS) regression model. Based on the comparison of the predictive regression curve, we can get that the SG smooth has a better reflection on the results. And when the method of the combination of Savitzky-Golay (SG) smoothing and the method of a single iterative Bagging partial least squares (IBPLS) is used to build the regression model, the gasoline model parameters RMSEP is 0.001 125 g·mL-1, R is 0.992 5; diesel model parameters RMSEP is 0.001 384 g·mL-1, R is 0.989 3.