Intention-to-treat analysis is an important aspect of randomized controlled trials of health care interventions. The concept is now widely accepted in theory, but not always implemented in practice. Failure to analyse by intention-to-treat can give misleading and indeed life-threatening interpretations. In some studies, a case is put for estimating the effect that would have been observed if all patients had received the allocated treatment. Situations where this is valid are rare, but an example is given of such an exceptional study. The relevance of the intention-to-treat concept is not always taken into account in qualitative research. Interviews with new mothers who delivered their babies at home in a hypothetical controlled trial of home versus hospital confinement would provide fascinating accounts of the pleasure of successful delivery at home. But by definition the interviews would exclude the hazard and drama of necessary transfers to hospital due to complications in late pregnancy and early labour. The intention-to-treat approach would avoid this bias.