Background Recent evidence suggests that elevated ambient temperatures may trigger preterm delivery. Since results from studies in temperate climates are inconclusive, we investigated the association between temperature and the risk of preterm birth in Flanders (Belgium). Methods We used data on 807 835 singleton deliveries (January 1998-July 2011). We combined a quasi-Poisson model with distributed lag non-linear models to allow for delayed and non-linear temperature effects, accounting for the daily pregnancies at risk and their gestational age distribution. Results For moderate heat (95th vs 50th centile) up to 1 day before delivery (lag 0-1), the risk of preterm birth increased by 8.5% (95% CI 2.4% to 15.0%) when minimum temperature increased from 8.3 degrees C to 16.3 degrees C and by 9.6% (95% CI 1.1% to 18.7%) when maximum temperature increased from 14.7 degrees C to 26.5 degrees C. Corresponding estimates for extreme heat (99th vs 50th centile) were 15.6% (95% CI 4.8% to 27.6%) for minimum temperature (19.0 degrees C vs 8.3 degrees C) and 14.5% (95% CI 0.5% to 30.6%) for maximum temperature (30.7 degrees C vs 14.7 degrees C). Despite the increased risk of preterm birth associated with cold at lag 2 (and lag 1 for minimum temperature), cumulative cold effects were small. The per cent change in preterm birth associated with moderate cold (5th vs 50th centile) up to 3 days before delivery (lag 0-3) was 2.1% (95% CI -4.1% to 8.7%) for minimum temperature (-2.0 degrees C vs 8.3 degrees C) and 0.6% (95% CI -7.3% to 9.2%) for maximum temperature (2.5 degrees C vs 14.7 degrees C). Conclusions Even in a temperate climate, ambient temperature may trigger preterm delivery, suggesting that pregnant women should avoid temperature extremes. ; This study was supported by the European Research Council (ERC-310898), ERA-NET FP7 ACCEPTED, and by Hasselt University Fund (BOF). The Study Centre for Perinatal Epidemiology is financed and commissioned by the Flemish Agency for Care and Health (Agentschap Zorg en Gezondheid). AG was supported by the UK Medical Research Council (grants ID: MR/M022625/1 and G1002296).