Ethiopian medicinal plants used for the treatment of a variety of ailments including infectious diseases were screened for activity against human immunodeficiency virus type 1 (HIV-1) and type 2 (HIV-2). Seventy-one polar and nonpolar extracts derived from 21 plants belonging to 14 families were tested for inhibition of viral replication using HIV-1 (III(B)) and HIV-2 (ROD) strains. Selective inhibition of viral growth was assessed by the simultaneous determination of the in vitro cytotoxicity of each of the extracts against MT-4 cells. Six extracts made from the root bark of Bersama abyssinica Fresen, the leaves of Combretum paniculatum Vent., and Dodonaea angustifolia L.f., and the stem bark of Ximenia americana L. displayed antiviral activity at concentrations that were nontoxic to MT-4 cells. The highest selective inhibition of HIV-1 replication was observed with the acetone fraction of C. paniculatum and the methanol fraction of D. angustifolia which showed selectivity indices (ratio of 50% cytotoxic concentration to 50% effective antiviral concentration) of 6.4 and 4.9, and afforded cell protection of viral induced cytopathic effect of 100% and 99%, respectively, when compared with control samples. The greatest degree of antiviral activity against HIV-2 was achieved with the acetone extract of C. paniculatum (EC(50): 3 microg/mL), which also showed the highest selectivity index (32). The 50% cytotoxic concentration ranged from 0.5 microg/mL for the hexane extract of D. angustifolia L.f., the most cytotoxic of the extracts tested, to >250 microg/mL for some extracts such as the methanol fraction of Alcea rosea L., the least toxic tested. Only the polar extracts that were obtained by extraction with hydroalcohol, methanol or acetone exhibited inhibition of viral growth at subtoxic concentrations. The results obtained in this study enable the selection of extracts which show some specificity of action and support the further investigation of these extracts for their potential as new lead antiretroviral compounds.