Modern traffic management centers often incorporate incident management support systems. One of the major components of these systems is an incident response module. Decisions on incident response are generally made on a prediction of how much impact an incident will have on a freeway in the sense of delays to drivers. The goal of this research was to build the framework for an algorithm that can more accurately predict delays due to incidents in real-time given a sufficient amount of data. In this research, many existing delay estimation methods were examined. These include heuristic models, queuing theory, shock-wave theory, time-series methods, and simulation. Most of these methods possessed drawbacks that preclude them from being used in a delay prediction algorithm. The type of model that was found to give the most flexibility while still being fast enough for on-line applications is a macroscopic freeway simulation model. A dynamic extended model for urban freeways was proposed and tested that proved to model urban freeway traffic phenomena very well. This dynamic model was taken as the basis for a new on-line delay forecasting system. This research provides an architecture for incorporating a macroscopic simulation with procedures that can be effectively used to provide forecasts of traffic conditions for an urban freeway corridor in real time. The input procedures combine information about freeway characteristics, incidents, and traffic flow conditions into a data package that can be efficiently run through the simulation program. Additional research performed includes a unique survey on the phenomenon of dynamic lane clearance. This phenomenon has a significant impact on delay prediction and was implemented into the architecture. The output from the simulation in terms of traffic flow and speed forecasts is flexible enough be used in many traffic management applications. When used with a real traffic data set, the simulation has shown to realistically predict future traffic conditions when given the proper input data. An open architecture was used in building the framework, leaving space for additional models that can improve accuracy. Additional research into these areas will further improve the algorithm, increasing both the stability and the accuracy of the forecasts. / Master of Science