We systematically reviewed the accuracy of serological tests for recent infections with HIV that have become widely used for measuring population patterns incidence of HIV. Across 13 different assays, sensitivity to detect recent infections ranged from 42-100% (median 89%). Specificity for detecting established infections was between 49.5% and 100% (median 86.8%) and was higher for infections of durations longer than 1 year (median 98%, range 31.5-100.0). For four different assays, comparisons were made between assay-derived population incidence estimates and a reference incidence estimate. The median percentage difference between the assay-derived incidence and reference incidence was 26.0%. Serological assays have reasonable sensitivity for the detection of recent infection with HIV, but are vulnerable to misclassifying established infections as recent-potentially leading to biases in incidence estimates. This conclusion is highly qualified by the apparent absence of a standardised approach to assay evaluation. There is an urgent need for an internationally agreed framework for evaluating and comparing these tests.