As part of a health impact evaluation of a water supply and sanitation project in a rural area of Bangladesh, diarrhoeal morbidity was recorded in children 0-4 years of age using weekly recall in household interviews, during the period March 1984 to December 1987. During the baseline year, 1984, the incidence rate of all diarrhoea episodes (3.8 episodes per child per year), and those defined as persistent, duration greater than 14 days (0.6 episodes per child per year), showed a similar age distribution, peaking in the 12-23 month age group. Sixteen per cent of all episodes were classified as persistent, and this proportion was greatest in the 0-5 month age group (25%). Children suffering at least one episode of persistent diarrhoea in 1984 also experienced a higher incidence of acute diarrhoea (less than = 14 days duration) than those suffering acute diarrhoea only (4.2 versus 3.7 episodes per child per year). Persistent diarrhoea showed a similar seasonal pattern to that of all episodes. Rates of abdominal pain, isolation of Shigella spp and a diagnosis of dysentery were significantly higher in persistent episodes than in acute episodes. Closer follow-up of children during 1986 and 1987, through the recording of all periods of absence of the child from the home, showed that overall diarrhoea incidence rates were little affected when absence was taken into account, but that the incidence of persistent diarrhoea and the proportion of episodes classified as persistent were significantly reduced. The implications of this methodological problem are discussed.