Spontaneous alpha activity clearly present in relaxed wakefulness with closed eyes, drowsiness period at sleep onset, and REM sleep was studied with spatial segmentation methods in order to determine if the brain activation state would be modulating the alpha spatial microstates composition and duration. These methods of spatial segmentation show some advantages: i) they extract topographic descriptors independent of the chosen reference (reference-free methods), and ii) they achieve spatial data reduction that are more data-driven than dipole source analysis. The results obtained with this study revealed that alpha activity presented a different spatio-temporal pattern of brain electric fields in each arousal state used in this study. These differences were reflected in a) the mean duration of alpha microstates (longer in relaxed wakefulness than in drowsy period and REM sleep), b) the number of brain microstates contained in one second (drowsiness showed more different microstates than did relaxed wakefulness and REM state), and c) the number of different classes (more abundant in drowsiness than in the rest of brain states). If we assume that longer segments of stable brain activity imply a lesser amount of different information to process (as reflected by a higher stability of the brain generator), whereas shorter segments imply a higher number of brain microstates caused by more different steps of information processing, it is possible that the alpha activity appearing in the sleep onset period could be indexing the hypnagogic imagery self-generated by the sleeping brain, and a phasic event in the case of REM sleep. Probably, REM-alpha bursts are associated with a brain microstate change (such as sleep spindles), as demonstrated by its phasic intrusion in a desynchronized background of brain activity. On the other hand, alpha rhythm could be the "baseline" of brain activity when the sensory inputs are minimum and the state is relaxed wakefulness.