A probabilistic fracture mechanics model was developed for determining the risk of fatigue failure of steel highway bridges. The model consists of: (1) Stochastic inputs for crack growth rate, fracture toughness, initial crack size, and load history; (2) a deterministic fracture mechanics calculation of crack growth; and (3) a Monte Carlo simulation to obtain the output variable, i.e., the fatigue life. The following effects on the risk of failure were examined: inspection interval, truck weight, truck traffic, system versus detail reliability, and length of service life extension. The model was applied to three bridges with cover-plated girders. It was found to predict well the short service life of the Yellow Mill Pond Bridge, Connecticut, in which the cover-plate end details began to fail 12 years after bridge opening. The model is particularly useful in determining the risk of extending the service life of a bridge beyond its intended design life.