OBJECTIVES: Administrative databases are increasingly being used in perinatal pharmacoepidemiology. Although such studies have increased statistical power, they are prone to biases due to lack of information on potential confounders. Hence, the objective was to quantify the degree of bias introduced by the lack of data on smoking status and maternal pre-pregnancy weight in the association between gestational exposure to antidepressants and the risk of low birth weight (LBW) infants. METHODS: The Quebec Pregnancy Registry was used to sample a cohort of women who had delivered a liveborn singleton, and answered a self-administered questionnaire on lifestyles during pregnancy and family history. The association between antidepressant use during pregnancy and the risk of LBW was calculated unadjusting for smoking status and maternal pre-pregnancy weight using the Registry data. The association was further estimated adjusting for smoking and maternal pre-pregnancy weight using the questionnaire data. Percent bias in the Registry estimate was calculated using Schneeweiss' method. RESULTS: Overall, the unadjusted estimate (Registry data) for the association between antidepressant use during pregnancy and the risk of LBW was 1.16 (p>.05). Adjusting for smoking status resulted in an estimate of 0.99 (p>0.05), and for maternal pre-pregnancy weight, 1.19 (p>0.05). The percent bias introduced by the non-adjustment for smoking and maternal pre-pregnancy weight using only the Registry data was 13%, and did not change the direction of the effect. CONCLUSIONS: Although data on potential confounders are often missing in perinatal pharmacoepidemiologic studies using administrative databases, the bias that results for the non-adjustment does not necessarily invalidate findings.