The knowledge of the brain processes underlying mental disorders has significantly increased during the last decades, but in spite of this very important research effort a biological marker is not available for such disorders. For example, neurophysiological techniques (EEG and MEG),have been broadly utilized in the investigation of the most important psychiatric syndromes such as schizophrenia, major depression, bipolar disorder or obsessive/compulsive disorder. The outcomes of some of those neurophysiological studies allowed the building of statistical models with very high sensitivity and specificity, although those models did not reach day-to-day clinical practice. A potential explanation for this situation is an inadequate analysis procedure which might be missing some important quantums of information contained in brain signals. In this vein, new methods of non-linear analysis have been proposed for the investigation of neurophysiological data. Particularly, the analysis of brain signals' complexity has been broadly utilized in the investigation of psychiatric disorders. Parameters of EEG-MEG complexity usually estimate the predictability of brain oscillations and/or the number of independent oscillators underlying the observed signals. More importantly, complexity parameters seem to be sensitive to the temporal components of brain activity, and therefore might reflect the dynamical nature of psychiatric disorders. This paper reviews some of the most relevant studies within this field, especially those focusing on the diagnosis, follow-up and prediction of response to treatment.