The present study examined the consistency of bioelectrical impedance analysis (BIA) and anthropometric measurements in body composition analysis in a field study of prepubertal children using a representative group of 2286 5-7-year-old children from Kiel, north-west Germany. Body composition was assessed using anthropometric measures (A; four skinfolds) and BIA. Various published algorithms (according to Lohmann (1986) and Deurenberg et al. (1990) for A, Kushner et al. (1992), Schaefer et al. (1994) and Wabitsch et al. (1996) for BIA and Goran et al. (1996) for a combined approach) were used to estimate body composition. Using A resulted in a sum of four skinfolds varying between age-dependent median values of 24.0 and 28.2 mm in boys and 30.5 and 33.3 mm in girls. When fat mass (FM) was calculated from A, age- and algorithm-dependent differences in median values were observed, with values varying between 8.5 and 14.6% for boys and 1.11 and 14.9% for girls. Using different algorithms (Lohmann (1986) v. Deurenberg et al. (1990)) only minor inconsistencies were observed. BIA-derived resistance index (height2/resistance) varied between 18.8 and 24.4 cm2/omh for boys and 17.1 and 19.0cm2/ohm for girls. Using four different algorithms to estimate FM from BIA data resulted in high intra-individual variances in percentage FM (from 13.8 to 33.4) as well as in the prevalence of overweight (from 14.7 to 98.4% for boys and from 42.3 to 98.5 % for girls). Data obtained using the different BIA algorithms showed some, or even marked, inconsistencies as well as systematic deviations (an overestimation of FM at low percentage FM, Schaefer et al. (1994) v. Wabitsch et al. (1996)). When comparing BIA with A, BIA systematically overestimated FM. The differences between the results were influenced by BMI, gender and height. Considerable inconsistencies were observed at low BMI (<10th percentile) for girls and for small children. Although the within-observer as well as between-observer CV for both techniques are acceptable, we recommend caution in relation to the algorithms used for data analysis. The use of an interchange table of percentage FM derived from different algorithms for different percentile groups of skinfold thicknesses is recommended.