Existing dynamic traffic assignment formulations predominantly assume the time-dependent O-D trip matrix and the time-dependent network configuration to be known a priori for the entire planning horizon. However, there is also a need to provide real-time path information to network users under ATIS/ATMS when unpredicted variations in O-D desires and/or network characteristics (e.g. capacity reduction on certain links due to incidents) occur. This paper presents a rolling horizon framework for addressing the real-time traffic assignment problem, where an ATIS/ATMS controller is assumed to have O-D desires up to the current time interval, and short-term and medium-term forecasts of future O-D desires. The assignment problem is solved in quasi-real time for a near-term future duration (or stage) to determine an optimal path assignment scheme for users entering the network in real-time for the short-term roll period. The resulting model is intricate due to the intertemporal dependencies characterizing this problem. Two formulations are discussed based on whether a capability to reroute vehicles en route exists. A rolling horizon solution procedure amenable to a quasi-real time implementation of a multiple user classes (MUC) time-dependent traffic assignment solution algorithm developed previously by the authors is described. Implementation issues are discussed from the perspective of ATIS/ATMS applications.