This report summarizes the results of a project to demonstrate a method to validate and calibrate a fatigue model. The project examined 30-day work histories of locomotive crews prior to 400 human factors accidents and 1000 nonhuman factors accidents. A biomathematical fatigue model estimated crew effectiveness (the inverse of fatigue) based entirely on work schedule information and opportunities to obtain sleep. A reliable linear relationship existed between crew effectiveness and the risk of a human factors accident (r = - 0.93); no such relationship was found for nonhuman factors accidents. This result satisfied the criteria for model validation. A reliable time of day variation occurred in human factors accidents (r = 0.71) but not in nonhuman factors accidents. The risk of human factors accidents was elevated at any effectiveness score below 90 and increased progressively with reduced effectiveness. At an effectiveness score ≤ 50, human factors accidents were 65 percent more likely than chance. Human factors accident risk increases reliably when effectiveness goes below 70, a value that is the rough equivalent of a 0.08 blood alcohol level or being awake for 21 hour following an 8-hour sleep period the previous night. Below an effectiveness score of 70, accident cause codes indicated the kinds of operator errors consistent with fatigue, confirming that the relationship between accident risk and effectiveness was meaningful.