Methanol gasoline is a new type of environmentally friendly diesel fuel, and the performance and quality of methanol-gasoline are based on the diesel methanol content. In this work, mid-infrared spectroscopy was successfully used to evaluate the methanol contents in methanol-gasoline samples with the aid of chemometric approaches. First, the mid-infrared spectral data obtained were pre-processed by smoothing standard normal, multiple scatter correction (MSC), baseline correction and normalization, and the partial least-square (PLS) quantitative calibration models were established, and the best pre-processed method was found. It was found that the PLS model pre-processed by MSC was much better than others, and the r and RMSEP evaluated were 0.918 and 2.107, respectively. In order to simplify the model and improve the prediction accuracy, uninformative variable elimination (UVE) was used to select the optimal wavelengths, and the experimental results showed that the prediction ability was greatly improved. Different quantitative calibration models by UVE selected wavelength, such as partial least-square (PLS), principal component regression (PCR) and least square-support vector machine (LS-SVM) for measuring methanol content were established and their prediction results were compared. It was found that the UVE-PLS model was much better than others, and the r and RMSEP evaluated were 0.923 and 2.075. It suggested that infrared spectroscopy in the detection of methanol content in the methanol gasoline is feasible and can bring good prediction results. UVE is an effective method for methanol gasoline in the infrared spectrum of band selection method, which is significant for the development of oil chemical industry.