A static model of undiagnosed and diagnosed HIV infections by year of infection and year of diagnosis was constructed to examine the impact of changes in HIV case-finding and HIV incidence on the proportion of late diagnoses. With no changes in HIV case-finding or incidence, the proportion of late diagnoses in the USA would remain stable at the 2010 level, 32.0%; with a 10% increase in HIV case-finding and no changes in HIV incidence, the estimated proportion of late diagnoses would steadily decrease to 28.1% in 2019; with a 5% annual increase in HIV incidence and no changes in case-finding, the proportion would decrease to 25.2% in 2019; with a 5% annual decrease in HIV incidence and no change in case-finding, the proportion would steadily increase to 33.2% in 2019; with a 10% increase in HIV case-finding, accompanied by a 5% annual decrease in HIV incidence, the proportion would decrease from 32.0% to 30.3% in 2011, and then steadily increase to 35.2% in 2019. In all five scenarios, the proportion of late diagnoses would remain stable after 2019. The stability of the proportion is explained by the definition of the measure itself, as both the numerator and denominator are affected by HIV case-finding making the measure less sensitive. For this reason, we should cautiously interpret the proportion of late diagnoses as a marker of the success or failure of expanding HIV testing programs.