In many recent studies on cognitive radio (CR) networks, the primary user activity is assumed to follow the Poisson traffic model with exponentially distributed interarrivals. The Poisson modeling may lead to cases where primary user activities are modeled as smooth and burst-free traffic. As a result, this may cause the cognitive radio users to miss some available but unutilized spectrum, leading to lower throughput and high false-alarm probabilities. The main contribution of this paper is to propose a novel model to parametrize the primary user traffic in a more efficient and accurate way in order to overcome the drawbacks of the Poisson modeling. The proposed model makes this possible by arranging the first-difference filtered and correlated primary user data into clusters. In this paper, a new metric called the Primary User Activity Index,, is introduced, which accounts for the relation between the cluster filter output and correlation statistics. The performance of the proposed model is evaluated by means of traffic estimation accuracy, false-alarm probabilities while keeping the detection probability of primary users at a constant value. Simulation results show that the appropriate selection of the Primary User Activity Index, higher primary-user detection accuracy, reduced false-alarm probabilities, and higher throughput can be achieved by the proposed model.