We consider the problem of quality-of-service (QoS) provisioning for multiple traffic classes in a MIMO wireless network. This QoS provisioning is posed as a radio resource management (RRM) problem at a wireless node (e.g., a wireless mesh router) with multiple antennas. We decompose this RRM problem into two tractable subproblems, namely, the antenna assignment and the admission control problems. The objective of antenna assignment is to minimize the weighted packet dropping probability for the different traffic classes under constrained packet delay. The objective of admission control is to maximize the revenue of the wireless node gained from the ongoing connections for different traffic classes under constrained connection blocking probability and average per-connection throughput. The decision of antenna assignment is made in a short-term basis (e.g., for every packet transmission interval) while that of admission control is made in a long-term basis (i.e., when a connection arrives). Constrained Markov decision process (CMDP) models are formulated to obtain the optimal decisions on antenna assignment and admission control. To provide efficient channel utilization, the RRM framework considers adaptive modulation at the physical layer which exploits channel state information. Performance evaluation results show that this joint antenna assignment and admission control framework can provide class-based service differentiation while satisfying both the connection-level and packet-level QoS requirements.