The ratios of circumferences (waist/hip, waist/thigh) have been proposed in lieu of skinfold measurements for studies of obesity and body fat distribution. The purpose of this paper is to present our experience in measuring a series of body circumferences and skinfolds in an obese population, and seek the relationship between these two kinds of variables as indicators of fatness and body fat distribution using canonical correlation analysis. With this method, weighted vectors of circumferences on the one hand and skinfolds on the other, are formed in such a way that the correlation between the two sets of variables is maximized. The weights (regression coefficients) and their signs help us select the best combination of circumferences which describe a component of centralized obesity. Our experience showed that 21 percent of women would have been excluded from this multivariate analysis due to skinfolds what could not be measured (mainly due to poor fold definition). Few men were so excluded (6 per cent), and almost all circumferences could be measured in both sexes. A first canonical correlation was substantial (0.84 in women, 0.89 in men) and appeared to relate to level of fatness. A second canonical correlation was moderate (0.59 in women, 0.42 in men) and statistically significant (P less than 0.01) in both sexes. Only in women was it independent of age, but in both sexes it reflected differences in central and peripheral (especially lower limb) fat. The simple waist/thigh ratio correlated well with the second canonical variate (0.79 in women, 0.67 in men).(ABSTRACT TRUNCATED AT 250 WORDS)