© 2017 Elsevier B.V. Due to the opportunistically resource usage of users in cognitive radio sensor networks (CRSNs), the availability of network resources is highly variable. Therefore, admission control is an essential mechanism to manage the traffic of cognitive radio users in order to satisfy the quality of service (QoS) requirements of applications. In this study, a connection admission control (CAC) mechanism is introduced to satisfy the requirements of delay sensitivity and power consumption awareness. This proposed mechanism is modeled through a semi Markov decision process (SMDP) and a linear programming problem is derived with the aim of obtaining the optimal policy to control the traffic of CRSNs and achieving maximum reward. The number of required channels at each network state is estimated through a graph coloring approach. An end to end delay constraint is defined for the optimization problem which is inspired from Kleinrock independence approximation. Furthermore, a power-aware weighting method is proposed for this mechanism. We conduct different simulation-based scenarios to investigate the performance of the proposed mechanism. The experimental results demonstrate the efficiency of this SMDP-based mechanism in comparison to the last CAC mechanism in CRSNs.