Bioelectrical impedance analysis (BIA) is an inexpensive, quick and non-invasive method to determine body composition. Equations used in BIA are typically derived in healthy individuals of European descent. BIA is specific to health status and ethnicity and may therefore provide inaccurate results in populations of different ethnic origin and health status. The aim of the present study was to test the validity of BIA in Ethiopian antiretroviral-naive HIV patients. BIA was validated against the 2H dilution technique by comparing fat-free mass (FFM) measured by the two methods using paired t tests and Bland-Altman plots. BIA was based on single frequency (50 kHz) whole-body measurements. Data were obtained at three health facilities in Jimma Zone, Oromia Region, South-West Ethiopia. Data from 281 HIV-infected participants were available. Two-thirds were female and the mean age was 32·7 (sd 8·6) years. Also, 46 % were underweight with a BMI below 18·5 kg/m2. There were no differences in FFM between the methods. Overall, BIA slightly underestimated FFM by 0·1 kg (-0·1, 95 % CI -0·3, 0·2 kg). The Bland-Altman plot indicated acceptable agreement with an upper limit of agreement of 4·5 kg and a lower limit of agreement of -4·6 kg, but with a small correlation between the mean difference and the average FFM. BIA slightly overestimated FFM at low values compared with the 2H dilution technique, while it slightly underestimated FFM at high values. In conclusion, BIA proved to be valid in this population and may therefore be useful for measuring body composition in routine practice in HIV-infected African individuals.