The knowledge of path marginal cost (PMC) is central to system-optimal dynamic traffic assignment (SO-DTA) problems. In this paper, we propose a method to approximate PMC in general networks when traffic dynamics are modeled by either the point-queue or the kinematic wave traffic flow model. This study examines in detail the flow interactions between downstream and upstream bottleneck links, and shows that the changes in cumulative flows on all the network links caused by an arbitrary flow perturbation can be computed. This offers a way to approximate PMC, which is incorporated in the solution of the least marginal cost problem, a central component of the path-based SO-DTA problem. The approximation scheme allows us to solve path-based SO-DTA problems for general networks with and without queue spillback and/or departure time choices. Numerical examples are provided to demonstrate the effectiveness of the proposed method, and the results show that the SO state produces considerably lower total network cost, shorter congestion duration, and smaller travel delay on bottleneck links than those of produced by the user-optimal state, particularly when the departure time choice is considered. © 2012 Elsevier Ltd.