Although EEG is intended to record cerebral activity, extra-cerebral activities from sites other than the brain are also recorded. Extraneous activities, known as artifacts, are generated by biological and non-biological sources. Knowing the difference between true cerebral activity and extraneous artifacts is critical in the interpretation of neonatal EEG. And, because artifacts can mimic true brain activity, responsibility lies on neurologists and anyone designing EEG-based systems to sift the "wheat from the chaff'. It is their duty to be able to recognize artifacts and to remove or reduce their influence so that effective detection of seizures can be achieved. We propose few artifact removal techniques and we evaluate their efficiency in enhancing the detection of neonatal seizures. Our detection method is based on the matching pursuit (MP) using a coherent time-frequency (TF) dictionary. Evaluation of 35 neonatal EEG records indicated an 18.3% increase in detection accuracy when artifact removal techniques were implemented.