Over the last few years, deep brain stimulation ( DBS) with targets such as the subthalamic nucleus or the pallidum were found to be beneficial in the treatment of Parkinson's disease and dystonia. The investigation of the mechanisms of action of DBS by recording concomitant neural activities in basal ganglia is hampered by the large stimulus artefacts ( SA). Approaches to remove the SA with conventional filters, or other conventional digital methods, are not always effective due to the significant overlap between the spectral contents of the neuronal signal and the SA. Thus, such approaches may produce a significant residual SA or alter the neuronal signal dynamics by removing its frequency contents. In this work, we propose a method based on an on-line SA template extraction and on the Ensemble empirical mode decomposition ( EEMD) to automatically detect and remove the dynamics of the SA without altering the embedded dynamics of the neuronal signal during stimulation. The results, based on real signals recorded in the subthalamic nucleus during Motor cortex stimulation ( MCS) experiments, show that this technique, which may be applied on-line, effectively identifies, separates and removes the SA, and uncovers neuronal potentials superimposed on the artefact.