Introduction The measurement of body composition is an important tool in the assessment of the nutritional status. 1,2 Historically, interest in body composition focuses on the measurement of body fat (for example in relation to obesity) or fat free mass (for example in relation to energy metabolism). There are a large number of techniques available for the assessment of body fat percentage (BF%), but many of these techniques are not applicable in epidemiological studies for various rea- sons. 3–5 Unfortunately, the most accurate techniques such as densitometry, isotope dilution or dual energy X-ray absorp- tiometry (DXA) are expensive and/or need specific skill of the operator and/or adequate cooperation of the subject. For the detection of the prevalence of obesity in a popu- lation or in obesity management programmes, predictive methods for BF% are adequate tools as they are less costly and relatively easy to apply to large population groups. One main disadvantage of prediction formulas is that their validity is only proven in the population in which they were developed. If they are to be used in other populations, it is important to validate the formulas in subsamples before they can be generally applied. Most prediction formulas are developed in Caucasian populations and their use and applicability in other ethnic groups is too often just assumed. There are indications that the validity of prediction equations is different among ethnic groups, 6–9 but this was not found in all studies. 10,11 Also, within one ethnic group differences in body parameters may exist which can lead to differences in the validity of predic- tive equations. 12 The aim of the current comparative study was to test the validity of a few predictive methods for BF% based on simple anthropometric measurements and based on im- Asia Pacific J Clin Nutr (2000) 9(2): 93–98 Original Article Correspondence address: Dr Mabel Deurenberg-Yap, Department of Nutrition, Ministry of Health, 3, Second Hospital Avenue, Singapore 168937. Tel: + 65 4353531; Fax: + 65 4383605 Email: Mabel_Yap@moh.gov.sg Accepted 7 February 2000 Prediction of percentage body fat from anthropometry and bioelectrical impedance in Singaporean and Beijing Chinese Paul Deurenberg 1,5 PhD, Mabel Deurenberg-Yap 2 MSc, Jingzhong Wang 3 BSc, Fu Po Lin 2 MSc and Gordon Schmidt 4 PhD 1 Department of Human Nutrition and Epidemiology, Wageningen Agricultural University, The Netherlands 2 Department of Nutrition, Ministry of Health, Singapore 3 Institute of Nutrition and Food Hygiene, Beijing 4 School of Physical Education, Nanyang Technological University, Singapore 5 Nutrition Consultant, Singapore Body composition was measured in 205 male and female Beijing Chinese and in 148 male and female Singaporean Chinese, age 34 (mean) (range 18–68) years and body mass index (BMI) 22.3 (15.9–38.5) kg/m 2 . In Beijing Siri’s two-compartment model based on densitometry was used as a reference technique and in Singapore Siri’s three-compartment model based on densitometry and deuterium oxide dilution was used. In addition, body composition was predicted using equations based on anthropometry and bioelectrical impedance developed in Caucasian populations. Percentage body fat (BF%) predicted from BMI was systematically underestimated by about 1% in Beijing Chinese and by about 3.5% in Singaporean Chinese. The difference in bias (measured minus predicted BF%) between the two population groups could be explained by differences in frame size. The Durnin and Womersley equations for BF% based on skinfold thickness predicted BF% in the male and female Chinese groups adequately, with only a slight (less than 1% body fat) and not significant bias. The prediction of BF% based on the waist circumference (Lean’s formula) resulted in an unbiased estimate of BF% in females (bias about 1% body fat), whereas in males the formula systematically underestimated BF% by 3.5–5%. Bioelectrical impedance underestimated BF% systematically by 3%, in males and females to about the same extent. The bias of all prediction formulas was positively correlated with the level of body fatness and, except for impedance, also negatively correlated with age. The negative association of the bias with age indicates that the age-related increase in body fatness is lower in Chinese than in Caucasians. It can be concluded of the studied prediction techniques that only the skinfold methodology using the equations of Durnin and Womersley give valid mean estimates for both Chinese males and females. The other techniques require the development of population-specific prediction formula. Key words: anthropometry and bioelectrical impedance, body composition, body mass index, percentage body fat. 93