Background: Current potent antiretroviral therapy (using a protease inhibitor and two nucleoside reverse transcriptase inhibitors) produces a durable suppression of HIV replication in less than 75% of treated patients. Identification of predictors of successful therapy might allow the development of improved strategies to increase response rates. Methods: We analyzed retrospectively the results from a multicenter, randomized, double-blind Phase III study of combination anti-HIV therapy with nelfinavir, zidovudine, and lamivudine to evaluate the relationship between virological response over 48 weeks of treatment to variables which could potentially serve as early predictors of long-term response. The goal was to produce long-term suppression of viral load to sensitive (< 400 copies HIV RNA/ml) and ultrasensitive (< 50 copies HIV RNA/ml) limits of quantification with the Amplicor PCR assay. Findings: Baseline viral load, the change in viral load over the first 4 weeks of treatment, the 2 h post-dose nelfinavir levels and the time to respond to HIV RNA levels of < 400 copies/ml and < 50 copies/ml have the best predictive value in determining response and response duration. Patients who achieved very low viral nadirs (< 50 copies HIV RNA/ml) had significantly longer responses than those who achieved nadirs of 50-400 copies HIV RNA/ml. Interpretation: Parameters that can be measured easily at baseline or early after therapy is started can identify patients at high risk of failure with standard treatment. Such patients may be candidates for more aggressive therapy or for alternative strategies designed to improve outcome, in addition, these results support the use of ultra-sensitive HIV RNA assays to predict long-term outcome of anti-HIV therapy.