BACKGROUND: Biased enrollment and attrition compromise the power of clinical trials and limit generalizability of findings. We identify predictors of enrollment and retention for HIV-discordant couples enrolled in prospective studies in Zambia. PRINCIPAL FINDINGS: A total of 1995 discordant couples were invited to enroll. Predictors of nonenrollment, loss to follow-up, and missed appointments were evaluated using multivariate models. MF couples were more likely to be eligible and to enroll and less likely to be lost to follow-up than FM couples. Substantial losses to follow-up occurred between testing and enrollment (21.3% of MF and 28.1% of FM) and between enrollment and the first follow-up visit (24.9% of MF and 30.5% of FM). Among MF and FM couples, residence far from the clinic, younger age, and women's age at first intercourse ≤17 years were predictive of attrition. No income, ≤2 lifetime sex partners, no history of sexually transmitted infection in women, and recent extramarital contact in their male partners predicted attrition in FM couples. CONCLUSIONS: Discordant couples are critical to observational studies and clinical trials to prevent male-to-female and female-to-male transmission. Retention biases must be taken into account during analysis. Run-in designs that delay randomization may improve retention in clinical trials. © 2008 Lippincott Williams & Wilkins, Inc.