The worldwide increasing prevalence of childhood overweight seems to be due to an increasing proportion of extremely high body mass index (BMI) values rather than to a shift of the entire BMI distribution. These findings might be attributed to incremental exposure to risk factors particularly affecting overweight children. To assess the possible differences in associations of several risk factors by subgroups of children's BMI distribution. We applied quantile regression to cross-sectional data on 9698 German preschoolers (5-6 years) collected in 1999 and 2002. Sex- and age-specific BMI standard deviation scores (BMI-SDSs) were used as the outcome variable, and maternal BMI, maternal smoking during pregnancy, exclusive formula feeding, child's weight gain from birth to 2 years of life and low parental education as explanatory variables. All risk factors except formula feeding contributed to a positive shift in mean BMI-SDS. The estimated effects of all risk factors on BMI-SDS were greatest for children with the highest BMI-SDS value. For example, high television (TV) viewing (>2 h day(-1)) had an effect of 0.46 (95% confidence interval (CI): 0.46, 0.46) SDS units on overweight children (90th percentile), but only a 0.22 (95% CI: 0.11, 0.33) SDS effect on normal-weight children (50th percentile). For well-known risk factors of childhood overweight, stronger associations in children with higher BMI values were observed. These findings might possibly help to explain the secular shift in the upper BMI percentiles in children.