Study objective - Retrospective infant feeding data are important to the study of child and adult health patterns. The accuracy of maternal recall of past infant feeding events was examined and specifically the infant's age when breast feeding was stopped and formula feeding and solid foods were introduced. Design and setting - The sample consisted of Bedouin Arab women (n = 318) living in the Negev in Israel who were a part of a larger cohort participating in a prospective study of infant health and who were delivered of their infants between July 1 and December 15, 1981. Data from interviews conducted 12 and 18 months postpartum were compared to the standard data collected six months postpartum. Main results - As length of recall increased there was a small increase in the mean difference, and its standard deviation, between the standard and recalled age when breast feeding was stopped and formula feeding and solid foods were started. Recall on formula feeding was less accurate than recall on solid foods and breast feeding. In particular, among those 61% reporting formula use at the six month interview, 51% did not recall introducing formula when interviewed at 18 months. The odds ratio (95% CI) of stunting versus normal length for age for formula fed versus breast fed infants based on recall data (OR = 2.07; 95% CI 0.82-5.22) differed only slightly from those based on the standard data (OR = 2.21; 95% CI 0.77-6.37). The accuracy of a mother's recall varied with her child's nutritional status at the time of the interview, but not with other sociodemographic, infant, or interviewer characteristics. Conclusions - Retrospective infant feeding data based on maternal recall of events up to 18 months in the past can be used with confidence in epidemiological studies. However, data on formula feeding may not be as accurate as data on breast feeding and solid food feeding, and accuracy may decrease as length of recall increases.