50these model components requires a set of parameters and inputs whose values must be defined or initialized before model execution. Various algorithms then tie together the demand-and-supply components to assign the dynamic demand to the network and determine the temporal propagation of flows. The resulting traffic conditions (including speeds, densities, travel times, and delays) may be used for a variety of planning and real-time management applications.DTA models differ in the mechanisms used to capture the timevarying nature of demand-and-supply processes and their interactions. Whereas various approaches to DTA have been developed, the current state of the art involves using sophisticated simulation techniques to model different transportation and traffic phenomena. DynaMIT (1) and Dynasmart (2) are examples of simulation-based DTA models that have been designed for a wide range of planning and network management applications. Planning applications are generally offline (using archived traffic data from the past), whereas network management applications may have to be performed online using real-time data transmitted by the network's surveillance system.Given the modeling complexities inherent to DTA systems, it is natural that they require a large number of inputs and parameters that must be customized for each application. These inputs and parameters, such as O-D flows, capacities, and route choice model parameters, must be identified so that the model's outputs reflect real-world conditions observed at the site of deployment. The advent of intelligent transportation systems provides rich sources of traffic data that can be used for this purpose: calibration involves estimating the DTA model's many inputs and parameters so that the gap between available traffic measurements and the model's outputs is minimized.This paper focuses on offline calibration of DTA models with archived traffic sensor data. The resulting estimates of model inputs and parameters are expected to represent historical traffic patterns. Once calibrated, these models may be used in offline studies or in online situations. If online calibration (with real-time sensor data) is desired, the historical estimates may serve as starting values.The number of calibration variables associated with a DTA model can be large. These variables can also be grouped into components, such as demand-and-supply parameters. Owing to the generally complex nature of the calibration problem, the various components of a DTA model typically have been calibrated separately. Often, a small set of parameters are adjusted manually in an attempt to improve on the default parameter settings (3-5). When optimization-based methods are applied, simplifications specific to a particular component or data type are used. O-D estimation, for example, uses the assignment matrix, a linear approximation of the relationship between O-D flows and traffic counts (6-9). Flows from multiple intervals are also estimated sequentially, because the more efficient simultaneous Advances in...