Heat transfer measurements and predictions are reported for a turbulent, separated duct flow past a wall-mounted two-dimensional rib. The computational results include predictions using the standard k–ε model, the algebraic-stress (A-S) functionalized k–ε model, and the nonlinear k–ε model of Speziale (1987). Three different prescriptions for the wall functions, WF I, WF II, and WF III given, respectively, by Launder and Spalding (1974), Chieng and Launder (1980), and Johnson and Launder (1982), are examined. The experiments include laser-Doppler flow measurements, temperature measurements, and local Nusselt number results. For WF I, the nonlinear model yielded improved predictions and displayed the most realistic predictions of the streamwise turbulence intensity and the mean streamwise velocities near the high-speed edge of the separated layer and downstream of reattachment. However, no significant improvements in the surface heat transfer predictions were obtained with the nonlinear model. With WF I and WF II, the models underpredicted the local Nusselt numbers and overpredicted the flow temperatures. With WF III, the predicted results agree with the experimental Nusselt numbers quite well up to reattachment, after which it substantially overpredicted the Nusselt numbers. The AS functionalized model using only the high Re formulation and curvature corrections in Cartesian coordinates improved the temperature predictions substantially, with the predicted flow temperatures agreeing quite well with the measured temperatures.