Selective study participation can theoretically lead to selection bias. We explored this issue in the context of a multicentre cohort study of socio-economic disparities in preterm birth. Women with singleton pregnancies were recruited from four large Montreal maternity hospitals and invited to return for an interview, vaginal examination and venepuncture at 24-26 weeks of gestation. We compared the observed preterm birth rate (ultrasound confirmed) among the 5146 cohort women to that expected based on all 108 724 Montreal Census Metropolitan Area (CMA) singleton births for 1998-2000. The observed preterm birth rate in the study cohort was 5.1%, compared with 6.3% in the CMA (P < 0.001) (unadjusted morbidity ratio [95% CI] = 0.80 [0.71, 0.90]). Within each stratum of maternal education and neighbourhood income (the latter based on postal code matched links to the 2001 Canadian census), cohort women had substantially lower rates of preterm birth than women from the CMA. No significant association between socio-economic status (SES) and preterm birth was observed in the study cohort, except among 'indicated' (non-spontaneous) cases. The association between neighbourhood income and preterm birth was biased to the null in the study cohort, with adjusted odds ratios in the poorest vs. richest quintiles of 1.01 [0.63, 1.64] in the cohort vs. 1.28 [1.18, 1.39] in the CMA, although no such bias was observed for the association with maternal education assessed at the individual level. We speculate that the lower-than-expected preterm birth rate and attenuated association between neighbourhood income and preterm birth may be related to selective participation by women more psychologically invested in their pregnancies. Investigators should consider the potential for biased associations in pregnancy/birth cohort studies, especially associations based on SES or race/ethnicity, and carry out sensitivity analyses to gauge their effects.