We propose in this study to compare the performances of two stochastic approaches (independent component analysis [ICA] and canonical correlation analysis [CCA]) to remove the muscular artefacts (EMG) from surface EEG signals in the context of epilepsy. The goal is to choose the method that better enhance the signals of interest (transient events called interictal spikes occur between seizures and background activity). In this paper, realistic EEG epileptic spikes are simulated from the activation of an epileptic patch. Real muscle artefacts and EEG background are then added to the simulated surface EEG. Such data allow us to quantify the performance of denoising methods since we have the "ground truth".