Body fat percentage (BF%) was measured in 298 Singaporean Chinese, Malay and Indian men and women using a chemical four-compartment model consisting of fat, water, protein and mineral (BF%4C). In addition, weight, height, skinfold thickness and segmental impedance (from hand to hand) was measured. Body fat percentage was predicted using prediction equations from the literature (for skinfolds BF%SKFD) and using the manufacturer's software for the hand-held impedance analyser (BF%IMP). The subjects ranged in age from 18-70 years and in body mass index from 16.0 to 40.2 kg/m2. Body fat ranged from 6.5 to 53.3%. The biases for skinfold prediction (BF%4C-BF%SKFD, mean +/- SD) were -0.4+/-3.9, 2.3+/-4.1 and 3.1+/-4.2 in Chinese, Malay and Indian women, respectively, the Chinese being different from the Malays and Indians. The differences were significant from zero (P < 0.05) in the Malays and Indians. For the men, the biases were 0.5+/-3.8, 0.0+/-4.8 and 0.9+/-4.0 in Chinese, Malays and Indians, respectively. These biases were not significantly different from zero and not different among the ethnic groups. The biases for hand-held impedance BF% were -0.7+/-4.5, 1.5+/-4.4 and 0.4+/-3.8 in Chinese, Malay and Indian women. These biases were not significantly different from zero but the bias in the Chinese was significantly different from the biases in the Malays and Indians. In the Chinese, Malay and Indian men, the biases of BF%IMP were 0.7+/-4.6, 1.9+/-4.8 and 2.0+/-4.4, respectively. These biases in Malay and Indian men were significantly different from zero and significantly different from the bias in Chinese men. The biases were correlated with level of body fat and age, and also with relative arm span (arm span/height) for impedance. After correction, the differences in bias among the ethnic groups disappeared. The study shows that the biases in predicted BF% differ between ethnic groups, differences that can be explained by differences in body composition and differences in body build. This information is important and should be taken into account when comparing body composition across ethnic groups using predictive methods.