Objective To examine the shape (functional form) of the association between the rate of gestational weight gain, pre-pregnancy body mass index (BMI), and preterm birth and its subtypes. Design Retrospective cohort study. Setting National reference obstetric centre in Lima, Peru. Population Pregnant women who delivered singleton babies during the period 2006-2009, resident in Lima, and beginning prenatal care at ≤12 weeks of gestation (n = 8964). Methods Data were collected from the centre database. The main analyses consisted of logistic regression with fractional polynomial modelling. Main outcome measures Preterm birth and its subtypes. Results Preterm birth occurred in 12.2% of women, being mostly idiopathic (85.7%). The rate of gestational weight gain was independently associated with preterm birth, and the shape of this association varied by pre-pregnancy BMI. In women who were underweight, the association was linear (per 0.1 kg/week increase) and protective (OR 0.88; 95% CI 0.82-1.00). In women of normal weight or who were overweight, the association was U-shaped: the odds of delivering preterm increased exponentially with rates <0.10 or >0.66 kg/week, and <0.04 or >0.50 kg/week, respectively. In women who were obese, the association was linear, but non-significant (OR 1.01; 95% CI 0.95-1.06). The association described for preterm birth closely resembled that of idiopathic preterm birth, although the latter was stronger. The rate of gestational weight gain was not associated with indicated preterm birth or preterm prelabour rupture of membranes. Conclusions In Peruvian pregnant women starting prenatal care at a;circ12 weeks of gestation, the rate of gestational weight gain is independently associated with preterm birth, mainly because of its association with idiopathic preterm birth, and the shape of both associations varies by pre-pregnancy BMI. RCOG.