The objectives of this paper were to: a) evaluate the accuracy and precision of previously published pediatric body composition prediction equations and b) develop additional prediction equations from a large, heterogeneous group of Caucasian (n = 133) and African-American (n = 69) children. The combined cohort of 202 children included a wide range of ages (4.0-10.9 y), weights (14.0-70.8 kg), fat mass (FM: 1.2-28.5 kg) and percent body fat (% body fat: 6.2-49.6%). Skinfold measurements were obtained using a Lange caliper and body fat was measured with a Lunar DPX-L densitometer. The previously published equations of Slaughter et al and Goran et al did not accurately predict body fat. The entire cohort was randomly divided into two sub-groups for purposes of deriving and cross-validating a new prediction equation. In stepwise regression analysis in the development group (n = 135), weight, triceps skinfold, gender, ethnicity and abdominal skinfold estimated FM measured by dual energy x-ray absorptiometry (DEXA) with a model R2 of 0.95. The new prediction equation was cross-validated in the control group (n = 67) and each ethnic and gender subgroup. We conclude that a) the equations of Slaughter et al and Goran et al did not accurately predict FM in a heterogeneous group of children and b) a new anthropometric prediction equation is proposed that may provide accurate estimates of FM in both Caucasian and African-American children aged 4-10 y with a wide range of FM and body composition.