Cyclostationary detection is considered as one of the major methods for spectrum sensing in cognitive radio. But due to cyclic frequency mismatch the performance degrades as the sensing time increases. In this paper we propose block based cyclostationary detection which overcomes the sensitivity of traditional cyclostationary detection to cyclic frequency mismatch. Theoretical analysis of block based cyclostationary detection is presented which shows that there is an optimal block length for which the performance is best. Simulation results corroborating the theoretical analysis are then presented. © 2013 IEEE.