EEG recordings are often contaminated by environmental noise and physiological factors commonly known as artifacts, which affect the quality of the signal. This paper presents an automatic method to classify between artifactual and neural components in EEG signals using an Independent Component Analysis (ICA) and a Support Vector Machine. With the resultant model, we obtained a classification accuracy of 95.6% validating the model over real data. The proposed methodology outperforms other approaches, as it achieves similar accuracy using exclusively ICA components as features.