We used signal detection theory to examine if grade crossing warning devices were effective because they increased drivers’ sensitivity to a train’s approach or because they encouraged drivers to stop. We estimated d' and β for eight warning devices using 2006 data from the Federal Railroad Administration’s Highway-Rail Grade Crossing Accident/Incident database and Highway-Rail Crossing Inventory. We also calculated a measure of warning device effectiveness by comparing the maximum likelihood of an accident at a grade crossing with its observed probability. The 2006 results were compared to an earlier analysis of 1986 data. The collective findings indicate that grade crossing warning devices are effective because they encourage drivers to stop. Warning device effectiveness improved over the years, as drivers behaved more conservatively. Sensitivity also increased. The current model is descriptive, but it provides a framework for understanding driver decision-making at grade crossings and for examining the impact of proposed countermeasures.