Nine databases were searched for studies on test performance in screening for oral cancer and precancer in primary care. Of 481 papers, full texts of 47 were scrutinised by two reviewers. Only prospective investigations of population screening involving examination of the oral mucosa, with gold standard verification (examination by an expert of subjects screened positive and at least a proportion of negatives), were selected. Seven papers describing eight studies were finally included (kappa=0.83). The weighted pooled value of sensitivity, from random effects meta-analysis, was 0.848 (95% CI 0.730, 0.919). The corresponding value for specificity was 0.965 (95% CI 0.930, 0.982). Main sources of clinical heterogeneity occurred between large house-to-house case finding programmes from SE Asia, utilising primary health workers, and smaller studies from England and Japan. Meta-analysis regression showed no difference (P=0.99) in the generally high level of discriminatory ability of the test between these two groups.