In this article, a Raman spectroscopy classification method combining the fluorescence background rejection based on graphitized carbon black (GCB) preprocessing and modified hierarchical clustering analysis (HCA) has been put forward, by which 39 fuel samples have been classified correctly and automatically into six types of 0# automobile diesel fuel, 0# general diesel fuel, 97# gasoline, 93# gasoline and 3# jet fuel. GCB preprocessing, during which 50mg GCB is used to filter 0.75 mL fuel sample once, has no influence on those samples which have no fluorescence background such as 3# jet fuels and gasolines and can reject effectively the weak fluorescence background of gasoline and automobile diesel fuel samples and strong fluorescence background of gasoline and general diesel fuel samples. Firstly, the Procrustes distance in the Procrustes analysis (PA) was adopted to measure the similarity between the samples for the classical HCA which was regarded as the modeling stage in the modified HCA algorithm. And the centroid vectors belonging to the different clusters were calculated according to the results of the HCA. Secondly, the types of the unknown test samples could be determined by calculating and comparing the Procrustes distances between the test samples and the centroid vectors of the clusters. The “leave-one-out” cross validation results based on the 39 samples belonging to 6 classes have shown that the GCB preprocessing is an effective fluorescence rejection method for Raman spectra of light fuels which can be applied to qualitative and quantitative analysis.