A　mental　stress　recognition　model　that　adapts　to　human　recognition　process　is　designed　after　analyzing　the　generation　of　mental　stress　in　this　paper.　Features　extraction　and　stress　quantification　are　achieved　by　the　model　with　a　stress　evaluation,　which　considers　affective　computing　theory,　based　on　the　Hidden　Markov　Model.　The　fusion　parameter　dataset　of　stress　is　built　with　features　acquired　by　physiological　parameters　of　stressed　individuals.　The　model　parameters　are　acquired　after　training　the　model　by　the　Baum-Welch　algorithm.　In　applying　the　model,　we　collected　17　training　samples　and　22　test　samples　from　the　volunteers＇　physiological　parameters　when　motivated　by　stress　sources.　The　stress　levels　of　all　samples　were　determined　by　the　questionnaire.　After　training　the　model,　the　accuracy　rate　of　the　model　for　the　test　sample　reached　96.4%.