Emotion recognition is essential for psychological and psychiatric applications and for improving the quality of human-machine interaction. Therefore, a simple and reliable method is needed to automatically assess the emotional state of a subject. This paper presents an application of clustering algorithms to feature spaces obtained from the acquired EEG of subjects performing a stress-inducing task. These features were obtained in three ways: using the EEG directly, using ICA to remove eye movement artifacts, and using EMD to extract data-driven modes present in the signals. From these features, we computed band-power features (BPFs) as well as pairwise phase-locking factors (PLFs), in a total of six different feature spaces. These six feature spaces are used as input to various clustering algorithms. The results of these clustering techniques show interesting phenomena, including prevalence for low numbers of clusters and the fact that clusters tend to be made of consecutive test lines.