This analysis of infant mortality in Bangladesh focuses on explaining death clustering within families, using prospective data from a rural region in Bangladesh, split into areas with and without extensive health services (the area covered by the International Centre for Diarrhoeal Disease Research and the comparison area, respectively). The modelling framework distinguishes between two explanations of death clustering: (observed and unobserved) heterogeneity across families and a causal 'scarring' effect of the death of one infant on the survival chances of the next to be born. Keeping observed and unobserved characteristics constant, we find scarring in the comparison area only. There the likelihood of infant death is about 29 per cent greater if the previous sibling died in infancy than otherwise. This effect mainly works through birth intervals: infant deaths are followed by shorter birth intervals, which increases the risk of infant death for the next child.