The present study was performed to develop regression-based prediction equations for fat mass by ultrasound in Japanese children and to investigate the validity of these equations. A total of 127 healthy Japanese pre-pubertal children aged 6–12 years were randomly separated into two groups: the model development group (fifty-four boys and forty-four girls) and the validation group (eighteen boys and eleven girls). Total body, trunk, arm and leg fat masses were initially determined by dual-energy X-ray absorptiometry (DXA, Delphi A-QDR whole-body scanner; Hologic, Inc., Bedford, MA, USA). Then, fat thickness was measured by B-mode ultrasound (5 MHz scanning head) at nine sites (arm: lateral forearm, anterior and posterior upper arm; trunk: abdomen and subscapular; leg: anterior and posterior thigh, anterior and posterior lower leg). Regression analyses were used to describe the relationships between the site-matched fat masses (total body, arm, trunk and leg) obtained by DXA and ultrasound in the development group. When these fat mass prediction equations were applied to the validation group, the measured total and regional fat mass was very similar to the predicted fat mass (mean difference calculated as predicted − measured fat mass ± 2 sd; total body 0·1 (sd 0·5) kg, arm 0·1 (sd 0·3) kg, trunk − 0·1 (sd 0·3) kg, leg 0·1 (sd 0·5) kg for boys; total body 0·5 (sd 1·3) kg, arm 0·0 (sd 0·3) kg, trunk 0·1 (sd 0·8) kg, leg 0·3 (sd 0·6) kg for girls), and the Bland–Altman analysis did not indicate a bias. These results suggest that ultrasound-derived prediction equations for boys and girls are useful for estimating total and regional fat mass.