Background Studies on the health benefits from breastfeeding often rely on maternal recall of breastfeeding. Although short-term maternal recall has been found to be quite accurate, less is known about long-term accuracy. The objective of this study was to assess the accuracy of long-term maternal recall of breastfeeding duration. Methods In a prospective study of pregnancy and birth outcome, detailed information on breastfeeding during the child’s first year of life was collected from a cohort of Norwegian women who gave birth in 1986–88. Among 374 of the participants, data on breastfeeding initiation and duration were compared to recalled data obtained from mailed questionnaires some 20 years later. Intraclass correlation coefficient (ICC), Bland-Altman plot, and Kappa statistics were used to assess the agreement between the two sources of data. Logistic regression was used to assess predictors of misreporting breastfeeding duration by more than one month. Results Recorded and recalled breastfeeding duration were strongly correlated (ICC=0.82, p < 0.001). Nearly two thirds of women recalled their breastfeeding to within one month. Recall data showed a modest median overestimation of about 2 weeks. There were no apparent systematic discrepancies between the two sources of information, but recall error was predicted by the age when infants were introduced to another kind of milk. Across categories of breastfeeding, the overall weighted Kappa statistic showed an almost perfect agreement (κ = 0.85, 95% confidence interval [CI] 0.82 – 0.88). Conclusion Breastfeeding duration was recalled quite accurately 20 years after mothers gave birth in a population where breastfeeding is common and its duration long.