The phenotype/genotype (PhenGen) open-label, randomized, multicenter study evaluated the genotype/virtual phenotype (vPt) and real phenotype (rPt) for choosing a new highly active antiretroviral therapy regimen at failure. Patients with a plasma viral load (pVL) between 2000 and 200,000 copies/mL and a CD4 cell count >200/microL, failing > or =2 regimens (<6 drugs), were randomized for vPt or rPt. Three hundred three patients were enrolled: 111 and 108 patients received a new treatment in the vPt and rPt arms, respectively. The 2 groups were comparable for baseline patient characteristics and treatment history. The new therapy was in agreement with expert advice in 58.5% of cases. After 6 months, no statistical differences were found in the mean absolute change from baseline CD4 cells (+55 and +46 cells/muL; P = 0.7), mean pVL log decrease (-1.35 and -1.37; P = 0.8), or proportion of patients with a pVL <400 copies/mL (54.8% in vPt arm and 52.6% in rPt arm; P = 0.9). At multivariate analysis, variables independently associated with failure of the new regimen were: pVL at baseline (odds ratio [OR] = 1.81; P < 0.021), number of nucleoside reverse transcriptase inhibitor-associated mutations (OR = 1.21; P = 0.001), number of protease mutations (OR = 1.15; P < 0.001), and recycling of indinavir (OR = 4.63; P = 0.019). Patients' adherence to the prescribed regimen (OR = 0.23; P < 0.001), number of active drugs in the new regimen (OR = 0.55; P = 0.001), and adherence to expert advice (OR = 0.37; P < 0.001) predicted virologic response. The vPt is as predictive of treatment outcome as the rPT. Use of expert advice significantly improved the response to therapy.