In this paper, we propose a channel reservation scheme for a cognitive radio network where a fraction of channels are strictly reserved for the Primary user (PU) transmissions only. The PU's and the SU's data packets arrive according to a Poisson process. We develop a probabilistic model for the PU occupancy and the probabilities of transmission with and without interference. Using this occupancy model we characterize the PU's and the SU's average achievable data rate when the PU faces spectrum sensing errors. We then show that using a reservation based scheme significantly improves the PU data rate at a marginal loss to the SUs data rate. We also derive the optimal number of channel to be reserved that maximizes sum rate. © 2019 IEEE.