A novel method was proposed to discriminate different kinds of spilled oil. The identification of the spilled oils has great significance to developing the treatment program and tracking the source. The present method adapts to Fourier transform NIR spectrophotometer to collect the spectral data of simulation gasoline, diesel fuel and kerosene oil spills. The Sparse Nonnegative Matrix Factorization algorithm was used to extract features. Through training with 210 samples and 5-fold cross-validation, the authors constructed the qualitatvie analysis model based on support vector machine. The authors also researched the effect of the number of features and sparseness factor. The proposed method has the identification capabilities with the accuracy of 97.78％ for 90 samples for validation. The present method of SNMF-SVM has a good identification effect and strong generalization ability, and can work as a new method for rapid identification of spilled oil.