The objectives of the present study were to identify predictors of serum retinol concentration as well as to assess the prevalence of low serum retinol concentration, in both the whole population after correcting for the effect of serum C-reactive protein (CRP) (using multiple categories), and the healthy subgroup. A cross-sectional study of 579 apparently healthy children, aged 3-7 years from a Dhaka slum, Bangladesh, was conducted. The effects of age, gender, serum CRP and alpha1-antichymotrypsin, reported morbidity (during the previous 2 weeks), Ascaris lumbricoides and Trichuris trichiura infections, parental education, wasting, stunting and underweight on serum retinol were estimated using multiple linear regression. The mean serum retinol concentration was 0.84 (sd 0.27) micromol/l. Elevated serum CRP levels, reported diarrhoea, reported nasal discharge and T. trichiura infection were negative predictors of serum retinol, whereas maternal education was a positive predictor. Compared with a serum CRP level of < 1 mg/l, CRP levels of 2 to < 5, 5 to < 10 and > or = 10 mg/l were associated with 0.12, 0.16 and 0.32 micromol/l lower serum retinol, respectively. The prevalence of low serum retinol (< 0.70 micromol/l) fell from 31.2 % to 15.6 % in the whole population, after correcting for the effect of CRP, and was 20.1 % in the healthy subgroup (CRP < 2 mg/l). The prevalence of low serum retinol was high but overestimated due to the effect of CRP. Interventions are needed to address low serum retinol in Bangladesh. Controlling diarrhoea, nasal discharge and T. trichiura infection and improving maternal education may be important interventions. The use of multiple categories of acute-phase proteins and cut-off values that indicate elevated levels need further research.