The services used by people with dementia and their carers were measured at three time points over 17 months. This analysis is unusual in that both informal care and formal inputs were costed. The costs estimates for informal inputs developed here may be applied to other data sets. Two hypotheses to explain the inter-relationship between informal and formal care inputs, substitution and supplementation, were compared in analysing the data. This paper explores the variations in costs according to the living arrangements of the people with dementia, by level of dependency and over time, and finds systematic differences. The costs of the inputs from co-resident carers consistently exceeded the costs of formal services, with informal care constituting up to 40 per cent of the total costs burden for dementia care. When informal inputs were included, non-domestic residential care emerged as less, not more, costly than care in the community. There was a shift in costs burden from health services to social services over time as more people moved into non-domestic settings. Multivariate analyses identified several predictors of informal and formal care costs: physical disability, level of cognitive impairment, living in non-domestic settings, and formal care provided. Non-domestic care predicted lower inputs of both formal and informal services. Dementia level was positively associated with informal inputs. Physical frailty was associated with more formal care, but less informal care. More formal service inputs predicted higher informal care inputs. There is evidence of supplementation of informal care by formal services in the early stages of care, followed by substitution as the person with dementia enters residential care.