Cognitive radio (CR) is a promising approach to improve the efficiency use of the wireless spectrum. One key element of this technology is spectrum sensing, which allows secondary users to detect the presence of licensed (primary) users. To this end, multiantenna spectrum sensing techniques have been proposed to detect the presence of a primary user based solely on the correlation structure of the signal received by a cognitive secondary receiver equipped with multiple antennas. Despite the numerous theoretical studies in the area of spectrum sensing, there exists a lack of experimental work evaluating the performance of these techniques in practice. In this paper, we test the performance of multiantenna Bayesian and generalized likelihood ratio test (GLRT) detectors on a cognitive radio platform. In comparison to one-shot GLRT detectors, the Bayesian detector is able to exploit past information from previous sensing periods, thus learning from the environment and improving its performance. Our cognitive platform is composed of Universal Software Radio Peripheral (USRP) nodes, that emulate the behavior of a single-antenna primary and a multiantenna cognitive receiver. Our measurements show that the Bayesian detector outperforms the GLRT detectors, in both stationary and nonstationary environments.