Retrospective reports of smoking in pregnancy are of importance for clinical or scientific purposes. Careful analyses of stability and accuracy of recalled behaviour are, therefore, needed. In 1998, the mothers of 2369 pre-teens born in Sweden retrospectively reported their smoking behaviour during the first trimester of the index pregnancy. We matched these reports with those recorded by midwives at the beginning of the index pregnancy, using information from the Swedish Medical Birth Registry. Using this registry as gold standard, the sensitivity and specificity of the retrospective reports containing any smoking were 83.9% and 92.8% respectively, but the sensitivity was low for daily smoking, 56.0%. Of the 222 discordant reports, 19.0% were due to mothers recalling daily smoking which was not reported at the time of pregnancy, and 42% were due to failure to recall smoking reported at the time of pregnancy, while the remaining 39% retrospectively reported occasional smoking, whereas they were registered as non-daily smokers when pregnant. Retrospective recall of pregnancy smoking is fairly stable over time.