Development of a robust technique for automatic detection of the epileptic seizures is an important goal in clinical neurosciences. In this paper a novel technique for detection of epileptic seizure based on constrained blind source separation (CBSS) has been developed. The BSS algorithm separates the electroencephalograms (EEGs) into their constituent independent components whereas imposing a constraint into the learning process highlights the epileptic source in the output of the separator. A reference signal (as a constraint) is modelled as a segment of a sine wave with a peak representing the seizure spike. The cycle frequency of the reference signal is estimated by looking at the spectrum of the separated EEG segment in the previous iteration. The results show an improvement in terms of computational speed in comparison with other systems.