Modeling　of　window　behavior　is　a　key　component　for　building　performance　simulation,　due　to　the　significant　impact　of　opening/closing　windows　on　indoor　environment　and　energy　consumption.　The　predictions　of　existing　models　cannot　well　reflect　actual　window　behavior,　the　prediction　accuracy　still　needs　to　be　improved.　The　Gauss　distribution　model　is　a　new　machine-learning　technique　which　has　achieved　successful　applications　in　many　fields　because　of　its　special　advantages　(i.e.　simple　structure,　strong　operability　and　flexible　nonparametric　inference　ability)　compared　to　existing　models.　This　paper　presents　results　from　a　study　using　the　Gauss　distribution　model　to　predict　window　behavior　in　office　building.　The　data　used　in　this　study　were　from　a　real　building　located　in　Beijing,　China,　and　covered　two　transitional　seasons　(from　October　1　to　November　15,　2014　and　from　March　15　to　May　16,　2015),　when　natural　ventilation　was　fully　applied.　When　modeling,　three　types　of　input　variables,　i.e.,　indoor　temperature,　outdoor　temperature　and　their　combination　were　used.　This　work　validates　the　importance　of　selecting　suitable　input　variables　when　developing　Gauss　distribution　model.　This　study　also　compared　the　prediction　performance　between　the　Gauss　distribution　modeling　approach　and　the　Logistic　regression　modeling　approach,　which　is　the　most　popular　method　used　to　model　occupant　window　behavior　in　buildings.　The　results　showed　that　Gauss　distribution　models　could　provide　higher　prediction　accuracy,　with　9.5%　higher　than　Logistic　regression　model　when　using　suitable　inputs.　This　paper　provided　a　novel　modeling　method　that　can　be　used　to　predict　window　states　more　accurately　in　office　buildings.