Cognitive networking deals with using cognition to the entire network protocol stack to achieve stack- wide, as well as network-wide performance goals; unlike cognitive radios that apply cognition only at the physical layer to overcome the problem of spectrum scarcity. Adding cognition to the existing Wireless Sensor Networks (WSNs) with a cognitive networking approach brings about many benefits. To the best of our knowledge, almost all the existing researches on the Cognitive Wireless Sensor Networks (CWSNs) have focused on spectrum allocation and interference reduction, which are related to the physical layer optimization. In this paper, an inference and learning model for CWSNs, named LA-CWSN, is proposed. This model uses learning automata to bring cognition to the entire network protocol stack, with the aim of providing end-to-end goal. Learning automata are assigned to the parameters of the important network protocols. Each automaton has a finite set of possible values of the corresponding parameter, and it tries to learn the best one, which maximize the network performance. Each node in the network has its own group of learning automata, which act independently, however all nodes receive the same feedbacks from the environment. To clarify the proposed model a traffic control scenario in WSN is considered. Using the network simulator ns-2.35, we test the proposed inference and learning model for traffic control in a WSN. The results show that learning automata approach works well to apply cognition in WSN