Multiple antenna is introduced into spectrum sensing in cognitive radios recently. However, conventional multiple antenna spectrum sensing schemes exploited only space diversity. In this paper, we propose a new multiple antenna sensing scheme based on space and time diversity (MASS-BSTD). First, the primary user signal to be sensed is over-sampled at each antenna, and signal samples collected at the same time instant from different antennas are stacked into a column vector. Second, each column vector is utilized to estimate space correlation matrix that exploits space diversity, and two consecutive column vectors are utilized to estimate time correlation matrix that exploits time diversity. Third, the estimated space correlation matrix and time correlation matrix are combined and analyzed using eigenvalue decomposition to reduce information redundancy of signals from multiple antennas. Lastly, the derived eigenvalues are utilized to construct the test statistic and sense the presence of the primary user signal. Since the proposed MASS-BSTD exploits both space diversity and time diversity, it achieves performance gain over the counterparts that only exploit space diversity. Furthermore, the proposed MASS-BSTD requires no prior information on the primary user, the channel between primary user transmitter and secondary user receiver, and is robust to noise uncertainty. Theoretical analysis and simulation results show that the proposed MASS-BSTD can sense the presence of primary user signal reliably.