A critical step for avian influenza viruses to infect human hosts and cause epidemics or pandemics is acquisition of the ability of the viral hemagglutinin (HA) to bind to human receptors. However, current global influenza surveillance does not monitor HA binding specificity due to a lack of rapid and reliable assays. Here we report a computational method that uses an effective scoring function to quantify HA-receptor binding activities with high accuracy and speed. Application of this method reveals receptor specificity changes and its temporal relationship with antigenicity changes during the evolution of human H3N2 viruses. The method predicts that two amino acid differences at 222 and 225 between HAs of A/Fujian/411/02 and A/Panama/2007/99 viruses account for their differences in binding to both avian and human receptors; this prediction was verified experimentally. The new computational method could provide an urgently needed tool for rapid and large-scale analysis of HA receptor specificities for global influenza surveillance.