The duration of lactational amenorrhoea, and infant feeding patterns and behaviour, were investigated in a sample of 97 mother–infant pairs living in a poor urban area of Dhaka, Bangladesh. A seven-hour time allocation method was used to determine the number of breast-feeding bouts and their duration. The seven-hour observation period was conducted on five occasions: at birth, 1 month, 4 months, 8 months and either 10 or 11 months. The median duration of lactational amenorrhoea was determined to be 24·07 weeks using survival analysis. Mothers who breast-fed their babies for longer and more frequently had, on average, a longer period of lactational amenorrhoea. There was no relationship between sociodemographic characteristics of the mother and duration of lactational amenorrhoea, nor was there any significant relationship between maternal anthropometry and birth weight of the baby and duration of lactational amenorrhoea, but there was a tendency for women with lower body mass index to have longer durations. Using the Cox proportional hazards model, the best predictor of duration of lactational amenorrhoea was the mean of months 0 and 1 durations of breast-feeding, adjusted for the mean frequencies for those months. The introduction of weaning food was also an important predictor.