For several years, Virginia Tech and other schools have measured the frequency and severity of head impacts sustained by collegiate American football players in real time using the Head Impact Telemetry (HIT) System of helmet-mounted accelerometers. In this study, data from 37,128 head impacts collected at Virginia Tech during games from 2006 to 2010 were analyzed. Peak head acceleration exceeded 100 g in 516 impacts, and the Head Injury Criterion (HIC) exceeded 200 in 468 impacts. Four instrumented players in the dataset sustained a concussion. These data were used to develop risk curves for concussion as a function of peak head acceleration and HIC. The validity of this biomechanical approach was assessed using epidemiological data on concussion incidence from other sources. Two specific aspects of concussion incidence were addressed: the variation by player position, and the frequency of repeat concussions. The HIT System data indicated that linemen sustained the highest overall number of head impacts, while skill positions sustained a higher number of more severe head impacts (peak acceleration > 100 g or HIC > 200). When weighted using injury risk curves, the HIT System data predicted a higher incidence of concussion in skill positions compared to linemen at rates that were in strong agreement with the epidemiological literature (Pearson's r = 0.72-0.87). The predicted rates of repeat concussions (21-39% over one season and 33-50% over five seasons) were somewhat higher than the ranges reported in the epidemiological literature. These analyses demonstrate that simple biomechanical parameters that can be measured by the HIT System possess a high level of power for predicting concussion.