Turbulent Prandtl number distributions are measured in a laboratory boundary layer flow with bed roughness, active blowing and sucking, and scalar injection near the bed. The distributions are significantly larger than unity, even at large distances from the wall, in apparent conflict with the Reynolds analogy. An analytical model is developed for the turbulent Prandtl number, formulated as the ratio of momentum and scalar mixing length distributions. The model is successful at predicting the measured turbulent Prandtl number behavior. Large deviations from unity are shown in this case to be consistent with measurable differences in the origins of the momentum and scalar mixing length distributions. Furthermore, these deviations are shown to be consistent with the Reynolds analogy when the definition of the turbulent Prandtl number is modified to include the effect of separate mixing length origin locations. The results indicate that the turbulent Prandtl number for flows over complex boundaries can be modeled based on simple knowledge of the geometric and kinematic nature of the momentum and scalar boundary conditions.