Résumé : In sub-Saharan Africa where weight loss is very difficult to estimate, cross-sectional anthropometric indicators could be useful to predict human immunodeficiency virus (HIV)-associated mortality. The study objective was to look for threshold values of baseline body mass index, arm muscle circumference, and fat mass to predict the risk of death in HIV-infected adults included in a 1996-1998 trial of early cotrimoxazole chemoprophylaxis in Abidjan, Côte d'Ivoire (COTRIMO-CI-ANRS 059 trial). The authors graphically determined if consecutive anthropometric categories with the closest hazards ratios of the risk of death could be clustered to obtain a unique threshold that distinctly separated two categories. When the threshold values were determined, the authors estimated the hazards ratio of mortality of this two-category model. A significant increase of mortality was observed for a body mass index of < or =20.3 in men (hazards ratio = 2.6; 95% confidence interval (CI): 1.4, 5.0) and of < or =18.5 in women (hazards ratio = 2.2; 95% CI: 1.05, 4.5) and for a fat mass of < or =6% in men (hazards ratio = 4.6; 95% CI: 2.3, 9.4) and of < or =18% in women (hazards ratio = 2.4; 95% CI: 1.2, 4.9). No simple threshold could be identified for arm muscle circumference. In Côte d'Ivoire where chemoprophylaxis of opportunistic infections has recently been recommended to be widely initiated on clinical criteria, such thresholds may help to screen patients with higher risks of mortality.