Through the analysis of known accelerants and residues, a trained neural network was developed for classifying fire debris samples. Three unknown items taken from actual fire debris that had contained gasoline, kerosene and diesel fuel were classified using this neural network. Diesel fuel residue was correctly identified every time. Inconsistent fingerprint patterns and incorrect classification by the neural network of the other two items may have been the result of variations in sample composition, possibly due to the effects of weathering or increased sample humidities. Sorbent sampling prior to aroma detection reduced these problems and allowed improved neural network classification of the kerosene and gasoline residues. Figures, tables, references