The most common method for producing regional or metropolitan area travel forecasts in the United States is to apply the following four modeling steps sequentially: trip generation, trip distribution, mode choice, and route assignment. This traditional four-step process passes output from one step to the next as input. While the process had produced forecast results sufficiently accurate for many types of long-range transportation planning, it is commonly found that some of the outputs of the process are not consistent with inputs to earlier steps. The research undertaken in this project focused on methods to ensure that link speeds used in each step of the travel forecasting process are consistent with the final speeds estimated in the final step of the process. A variety of methods for introducing 'feedback' into the process (reintroducing output of one step as input to a previous step) were explored and guidance was developed on when and how to incorporate feedback into the four-step modeling process. Four possible ways in which feedback can be provided in the four-step process and one additional way that feedback can be accomplished to the forecasting of land-use and activity patterns is illustrated.