The performance of an artificial neural network was evaluated as an alternative classification technique of autofluorescence spectra of oral leukoplakia, which may reflect the grade of tissue dysplasia. Twenty-two visible lesions of 21 patients suffering from oral leukoplakia and six locations on normal oral mucosa of volunteers were investigated with autofluorescence spectroscopy (420 nm excitation, 465-650 nm emission). Pre-scaled spectra were combined with the corresponding visual and histopathological classifications in order to train artificial neural networks. A trained network is mapping input spectra to tissue characteristics, which was evaluated using a blind set of spectra. Abnormal tissue could be distinguished from normal tissue by a neural network with a sensitivity of 86% and a specificity of 100%. Also, classifying either homogeneous or non-homogeneous tissue performed reasonably well. Weak or no correlation existed between spectral patterns and verrucous or erosive tissue or the grade of dysplasia, hyperplasia and hyperkeratosis.