In an Integrated Services Digital Network (ISDN) environment, users with widely differing characteristics, e.g., voice and interactive data, share system resources, such as time and bandwidth. To maintain efficient operation of the system, access to the network has to be closely controlled. We study the problem of dynamically determining the best access policy at a single node of the network. We model the network node by means of a two-dimensional, continuous-time Markov chain, that describes the message queue size and the number of accepted voice calls. Using linear and dynamic programming arguments, we show that for certain criteria, the optimal control policy has the form of a "switching curve".