Objectives: To examine relationships between the cost of hospital cancer care and the degree of specialisation of patients' doctors and hospitals as indicated by their annual case-loads. Methods: Three cohorts comprising 2294 patients with cancers of the pancreas, oesophagus and stomach in 29 acute National Health Service (NHS) hospitals in south-west England and South Wales were followed prospectively for a year. For each patient, prognostic variables, service activity and survival data were recorded, and the hospital cost of cancer care and cost per year survived were estimated. Costs were estimated from quantities of resource units provided and a single set of unit costs. Costs per day survived after presentation to hospital were calculated as estimates of cost-effectiveness. Linear regression analyses examined the effects on total costs, and on costs per day survived of doctors' and hospitals' annual patient volumes, adjusting for prognostic variables and treatments. Results: General ward care rather than specific treatments and investigations accounted for most of total costs. Costs per patient increased significantly with increasing doctor volumes. After adjustment for prognosis and treatments, however, cost-volume relationships were U-shaped, reflecting more active intervention by higher volume doctors, along with little activity and long stays among patients of lower volume doctors. Cost per day survived also had U-shaped relationships with doctor volumes. Regression models using continuous rather than categorical volume terms fitted the data best. Conclusions: Doctors' specialisation is at least as important for efficiency and effectiveness as hospitals' specialisation. Cost and cost per day survived increased and then decreased with increasing doctor volumes, highlighting the need for critical attention to costs as services become highly specialised. Specialisation occurs along a continuum, with no clear volume threshold effects on survival or costs.