The problem of recall bias when reporting breast feeding duration is investigated. Data are presented from the follow-up of over 1,000 children from a birth cohort (1982) in southern Brazil, when they were on average 11, 23, and 47 months of age. Compared with the response given at 11 months of age, 24% of mothers misclassified the duration (grouped into 3-month categories) at age 23 months and 30% at age 47 months. Women who were richer and/or better educated were significantly more likely to report longer durations, while those poorer and less educated did not tend to misclassify more in one direction than in the other.