Since the mid-1990s, the number of studies linking air pollutants to low birthweight, small for gestational age, preterm birth and cardiac birth defects has grown steadily. This MiniReview (in conjunction with the May 2007 International Conference on Foetal Programming and Development Toxicity) highlights key methodological issues surrounding this research area, based on our experiences in Southern California. All 'criteria' air pollutants have been linked to birth outcomes. Our own studies found most consistent associations for carbon monoxide and particles. Traffic exhaust toxins are possible causative agents, but air monitoring data relied on by almost all existing studies inadequately capture their intracommunity variability in concentrations. Exposure assessment might be improved by biomarkers and land use-based regression modelling or information on time-activity patterns. Foetal development provides unique opportunities to study exposures acting during narrow susceptibility windows. A time-series approach by design controls for confounders that do not vary temporally but can only address short-term acute effects. Studies employing spatial or medium-term (trimester-specific) temporal contrasts may be more susceptible to residual confounding, and studies adjusting only for risk factors recorded on birth certificates have been criticized. Findings from our recent study in Southern California indicate that air pollution effect estimates are not markedly influenced by risk factors not provided on birth certificates. Yet, studies collecting detailed risk factor information in other geographic regions may be needed to further evaluate the extent of residual confounding in record-based analyses. Investigating biological mechanisms (e.g. using ultrasound measurements and biomarkers for hypothesized pathways) is an important remaining issue.