Sublethal heating can increase subsequent thermal resistance of bacteria, which may compromise the validity of thermal process validations for slow-roasted meats. Therefore, this research evaluated the accuracy of a traditional log-linear inactivation model, developed via prior laboratory-scale isothermal tests, and a novel path-dependent model accounting for sublethal injury, applied to pilot-scale slow cooking of whole-muscle roasts. Irradiated turkey breasts, beef rounds, and pork loins were inoculated with an eight-serovar Salmonella cocktail via vacuum tumble marination in a salt-phosphate marinade. The resulting initial Salmonella population in the geometric center (core) was 7.0, 6.3, and 6.3 log CFU/g for turkey, beef, and pork, respectively. Seven different cooking schedules representing industry practices were evaluated in a pilot-scale, moist-air convection oven. Core temperatures recorded during cooking were used to calculate lethality real-time via the log-linear model. The path-dependent model reduced the bias (mean residual) and root mean square error by 4.24 and 4.60 log CFU/g respectively, in turkey; however, the new model did not reduce the prediction error in beef or pork. Overall, results demonstrated that slow-cooked roasts, processed to a computed lethality at or near that required by the regulatory performance standards, as calculated with a state-dependent model, may be underprocessed.