Determining the predictors of serum retinol at mid-pregnancy is relevant for planning interventions aimed at improving vitamin A status of pregnant women and their offspring. This prospective study assessed predictors of serum retinol at the beginning of the third trimester of pregnancy. We enrolled 442 pregnant women living in the urban area of Cruzeiro do Sul, Western Brazilian Amazon. Demographic, socio-economic, environmental and clinical characteristics as well as obstetric history, anthropometric, dietary and biochemical data, including serum retinol, were gathered between 16 and 20 gestational weeks. Serum retinol also measured at the beginning of the third trimester of pregnancy (approximately 28 gestational weeks) was the outcome of interest. Multiple linear regression models were used to evaluate associations with the outcome. Overall, the following variables explained serum retinol at the beginning of the third trimester of pregnancy in the adjusted model (R2 = 11·1 %): seasonality (winter season – November to April; β=0·134; 95 % CI 0·063, 0·206), weekly consumption of Amazonian fruits (β=0·087; 95 % CI 0·012, 0·162) and retinol concentrations between 16 and 20 gestational weeks (β=0·045; 95 % CI 0·016, 0·074) were positively associated, whereas having a smoker in the house was negatively associated (β=–0·087; 95 % CI: –0·166, –0·009). Consumption of pro-vitamin A-rich fruits by pregnant women should be encouraged. Passive smoking may play a role in decreasing vitamin A status as a proxy of smoking exposure during pregnancy.