It is generally agreed that there is a need for a routine, non-invasive screening procedure for oral cancer particularly of high risk groups. Refinements in oral exfoliative cytology now make this technique worthy of consideration for such screening. This study assesses the utility of monitoring cytokeratin expression in smears of oral cancer in comparison with assessing the keratin expression in corresponding biopsies. Smears and biopsies were taken from 34 patients with oral cancer. A panel of antibodies, CAM5.2, LH1, AE8, LP2K and LH8 recognising keratins 8, 10, 13, 19 and a basal cell marker respectively were employed. Keratins were identified using a standard immunocytochemical technique (Vectastain) and assessed on a 3 point scale, for both smears and biopsies. The vast majority of rumours were well differentiated. No particular keratin profile seen within the smear was associated with any particular state of differentiation. Although the sensitivity of K19 was greatest, its specificity was poor. The keratin antibodies with the best positive predictive value were CAM5.2 (K8) and the marker of the basal cell phenotype, LH8. The combination of down regulation of the secondary differentiation markers (K13, K10) coupled with 'simple' keratin expression (K8, K19) would seem to be the most consistent profile. We conclude that for exfoliative cytological screening to be of value as a diagnostic test it remains necessary to employ assays using more than one antikeratin antibody.