Serological assays are increasingly being used to measure HIV incidence in cross-sectional studies, but their specificity to determine incident infections remains problematic. We estimated the specificity of the BED assay in a cohort of long-term HIV-infected adults before and during antiretroviral treatment (ART) and evaluated an HIV avidity assay to detect BED-based false-recent results. We used the BED assay to test stored specimens from known long-term HIV-1-infected adult Ugandans before and at 3, 12, and 24 months after ART initiation. We evaluated the frequency of false-recent classifications by ART status and CD4(+) T(+) cell count. Specimens classified as BED false-recent were further tested with an avidity assay. In all, 950 blood specimens from 253 adults were tested with the BED assay. Of these, 149 (15.7%) specimens tested false-recent and 64 (24.9%) individuals tested false-recent at least once. Among all specimens tested, the proportion of false-recent rose with increasing CD4(+) cell count (<250 cells/μl: 11.3%, 250-499: 17.8%, ≥500: 21.4%; p for trend=0.002). Of 197 persons with all four BED results available, 75.6% were classified as long-term infected throughout and 8.1% as false-recent throughout; the remainder changed classification once (12.2%) or twice (4.1%). Of 105 false-recent specimens retested with the avidity assay, 101 (96.2%) were correctly classified as "long-term." The BED assay's specificity varied with CD4(+) cell count and use of ART. Knowledge of these parameters for blood samples could improve incidence estimates using the BED assay. The additional use of an avidity assay may help to minimize the proportion of BED false-recent specimens.