Artificial intelligence is being used increasingly as an aid to diagnosis in medicine. The purpose of this study was to evaluate the ability of a neural network to predict the likelihood of an individual having a malignant or potentially malignant oral lesion based on knowledge of their risk habits. Performance of the network was compared with a group of dental screeners in a screening programme involving 2027 adults. The screening performance was measured in terms of sensitivity, specificity and likelihood ratios. All subjects were examined independently by a dental screener and a specialist, who provided a definitive diagnosis, or 'gold standard', for each individual. All subjects also completed an interview questionnaire regarding personal details, dental attendance and smoking and drinking habits. The neural network was trained on 1662 of the screened population using ten input variables derived from the questionnaire along with the outcome of the specialist's diagnosis. Following training, the network was asked to classify the remaining unseen proportion (365 individuals) of the screened population as positive or negative for the presence of cancer or precancer. The overall sensitivity and specificity of the dentists were 0.74 [95% confidence interval (CI), 0.62-0.86] and 0.99 (95% CI, 0.985-0.994) respectively compared with 0.80 (99% CI, 0.55-1.00) and 0.77 (95% CI, 0.73-0.81) for the neural network. In view of the potential costs involved in implementing a screening programme, this neural network may be of value for the identification of individuals with a high risk of oral cancer or precancer for further clinical examination or health education.