Herein, we consider the problem of detecting primary users’ signals in the presence of noise correlation, which may arise due to imperfections in filtering and oversampling operations in a Cognitive Radio (CR) receiver. In this context, we study a Maximum Eigenvalue (ME) detection technique using recent results from Random Matrix Theory (RMT) for characterizing the distribution of the maximum eigenvalue of a class of sample covariance matrices. Subsequently, we derive a theoretical expression for a sensing threshold as a function of the probability of false alarm and evaluate the sensing performance in terms of probability of correct decision. It is shown that the proposed approach significantly improves the sensing performance of the ME detector in correlated noise scenarios.