Background: Accurate estimates of HIV incidence are needed to assess epidemics, calibrate models, and design and evaluate interventions. The crosssectional use of biomarker-based Tests for Recent HIV Infection (TRIs) in principle offers affordable, low-bias options for incidence estimation. To date, there has been no independent, directly comparative benchmarking of candidate TRIs. Methodology: A repository was assembled comprising replicate plasma samples from 5641 specimens representing 2007 subjects from studies in Africa, Brazil and the United States, with suitably characterized longitudinal subject information (estimated exposure dates, detailed treatment and clinical history, viral load, CD4 counts). Five TRIs (BED, Limiting Antigen (LAg), Detuned Vitros, Vitros Avidity and Biorad Avidity), previously assessed on a 250 sample ‘Qualification Panel’, were evaluated using a clade diverse 2500 member ‘Evaluation Panel’. Using developers’ previously published recent/non-recent discrimination criteria, frequency estimation and regression yielded estimates of the Mean Duration of Recent Infection (MDRI – mean time, within one year post estimated exposure, that subjects return recent results), False-Recent Rate (FRR – proportion of recent test results among patients at least one year post infection) and assay reproducibility (CoV – coefficient of variation of assay reading). FRR was calculated by time since infection for untreated patients, and separately by ARV treatment status. Results: Table 1 shows primary performance characteristics of the incidence assays. FRR is reported by treatment status and in a hypothetical population with 30% of positives ART suppressed. Treated versus untreated FRR differences all have p<0.001. Conclusions: This is the first independent evaluation of predicate incidence assays. According to developers’ published protocols, none appears suitable, in stand-alone form, as a widely applicable incidence surveillance tool. As viral suppression drives false recent results, optimized use of serologic assays, using low viral load as indicative of non-recent infection, could yield practical TRIs. Work is on-going to 1) evaluate additional assays, 2) optimise multiple biomarker algorithms, 3) estimate FRR with realistic population representative distributions of clinical stages, and 4) maintain and expand the repository to support new work including biomarker discovery.