Pancreatic cancer is the fifth most common cause of cancer death in the western world and the prognosis for unresectable disease remains poor. Recent advances in conventional chemotherapy and the development of novel 'molecular' treatment strategies with different toxicity profiles warrant investigation as combination treatment strategies. This randomised study in pancreatic cancer compares marimastat (orally administered matrix metalloproteinase inhibitor) in combination with gemcitabine to gemcitabine alone. Two hundred and thirty-nine patients with unresectable pancreatic cancer were randomised to receive gemcitabine (1000 mg m(-2)) in combination with either marimastat or placebo. The primary end-point was survival. Objective tumour response and duration of response, time to treatment failure and disease progression, quality of life and safety were also assessed. There was no significant difference in survival between gemcitabine and marimastat and gemcitabine and placebo (P=0.95 log-rank test). Median survival times were 165.5 and 164 days and 1-year survival was 18% and 17% respectively. There were no significant differences in overall response rates (11 and 16% respectively), progression-free survival (P=0.68 log-rank test) or time to treatment failure (P=0.70 log-rank test) between the treatment arms. The gemcitabine and marimastat combination was well tolerated with only 2.5% of patients withdrawn due to presumed marimastat toxicity. Grade 3 or 4 musculoskeletal toxicities were reported in only 4% of the marimastat treated patients, although 59% of marimastat treated patients reported some musculoskeletal events. The results of this study provide no evidence to support a combination of marimastat with gemcitabine in patients with advanced pancreatic cancer. The combination of marimastat with gemcitabine was well tolerated. Further studies of marimastat as a maintenance treatment following a response or stable disease on gemcitabine may be justified.