Whereas over the last decade epidemiologic studies on exocrine pancreatic cancer (EPC) continued to show a remarkable heterogeneity in diagnostic criteria applied to define caseness, the actual magnitude and consequences of misclassification remain largely unexplored. The objectives were: (1) to estimate the degree of certainty with which cases of EPC are diagnosed in the two participating hospitals (to this end a diagnostic certainty classification (DCC) was developed; (2) to test whether characteristics of cases differed by degree of diagnostic certainty; and (3) to assess what influence different definitions of case might have on risk estimates for tobacco and alcohol. All cases with a discharge diagnosis of EPC who attended at the Hospital del Mar between 1980-90 and at the Hospital Son Dureta between 1983-90 were identified through their respective tumor registries, and their clinical records were reviewed. Only 52% of 140 cases were classified in the group with a higher probability of EPC (group H). Diagnostic certainty appeared somewhat greater among women (age-adjusted odds ratio [ORa] 1.60, p = 0.18). Group H showed a higher proportion of cases with an interval from first symptom to diagnosis < or = 1 month (ORa = 2.38, p < 0.05) and the proportion of adenocarcinomas was slightly higher than in less certain cases (group L) (p = 0.051). A radical treatment was exclusively attempted in group H (p < 0.001). DCC cut-off points had a significant effect on the proportion of smokers and of alcohol drinkers, as well as on the percent of cases with pathological (cytohistological) confirmation. The proportion of cases unlikely to be of pancreatic origin in spite of having pathological confirmation was high enough to cause significant misclassification bias. Because past exposure to certain risk factors may differ among cases with different diagnostic certainty, we suggest to initially include in the case group patients who in spite of lacking pathological confirmation have strong clinical evidence supporting the diagnosis of EPC; subsequently, risk estimates should be computed across strata of diagnostic certainty to assess whether heterogeneity exists. In exocrine pancreatic cancer the impact of misclassification of disease status upon etiologic and prognostic estimates deserves at least as much attention as misclassification of exposure.