This　paper　proposes　a　multilevel　security　model　based　on　trustworthy　state,　which　introduces　the　concept　of　reliability　and　the　function　of　integrity　measurement.　It　enforces　diverse　security　rules　to　different　kinds　of　objects.　To　prevent　running　intrusions,　the　access　ranges　of　subjects　are　adjusted　according　to　their　reliabilities.　The　formal　description　of　the　model　and　the　transform　of　its　trustworthy　states　are　given　in　the　paper.　It　is　proved　that　the　system　remains　in　a　secure　state　after　performing　the　security　rules　of　the　new　model.　Moreover,　its　realizing　framework　in　operating　system　kernel　is　described,　and　its　performance　is　analyzed.