Although inthe present systems of emotion recognition are widely used in the phase of cognitive behavioral therapy, computer human interface, techniques of emotional management, the techniques developed processing even not reach percentages higher confidence due to problems at the same signals, such as filtering processes, devices, systems characterization not determine with certainty the emotional content present in these signals or classification systems that do not perform optimally in their task. Likewise the fundamental idea of each of the techniques with their pros and cons, in order to establish a comparison frame and thus to find a combination of techniques which offer the least error in recognizing emotion described present from a physiological signal. Based on the problems presented by the emotion recognition systems, requires a system capable of performing this task robustly implement techniques that are able to quantify the information in the different physiological signals studied. Therefore a methodology for the recognition of anxiety states proposed by multimodal analysis techniques and machine learning in physiological signals in order to apply this methodology in psychological treatment for anxiety management.