The results of 407 secretin-pancreozymin tests were analysed by a variety of statistical methods, in an attempt to optimize diagnostic yield. The best diagnostic yield accrued from selection of a point corresponding to 95% specificity and 60% sensitivity on each of two virtually superimposable receiver-operator curves--using either bicarbonate output or a discriminant function derived from multivariate analysis--and the anticipated yield approximated to that realised in a further prospective series of 150 cases. At a 25% hypothetical local prevalence of chronic pancreatic disease (including chronic pancreatitis and pancreatic cancer), the positive predictive value was 80%, the negative predictive value 88% and the efficiency was 86%--values that are not dissimilar to those reported in a study in which both hormones were given simultaneously by constant intravenous infusion for 105 min with multivariate analysis of the results. We conclude that (a) measurement of bicarbonate output in 30 min after an appropriate dose of secretin given as a bolus injection yields results that are comparable to those obtained when secretin and pancreozymin are given by constant intravenous infusion in doses to evoke maximal secretory responses; and (b) the yield of hormone tests using duodenal intubation is far from ideal.