Window　opening　behavior　in　maternity　hospital　wards　is　the　result　of　collective　actions　to　create　a　healthy　and　comfortable　recovery　area　for　the　patients　and　it　is　a　main　strategy　of　natural　ventilation.　Since　many　rooms　in　the　hospital　are　occupied　24　h　a　day,　which　could　lead　to　different　window　operation　characteristics　from　other　building　types.　However,　the　global　understanding　of　window　opening　behavior　at　maternity　hospitals　is　very　limited　as　little　research　has　been　undertaken.　In　order　to　fully　understand　the　behavioral　characteristic　and　improve　the　prediction　accuracy,　monitoring　data　from　a　maternity　hospital　in　Beijing　during　summer　were　analyzed　and　modeled　by　statistical　method　and　random　forest　(RF)　algorithm　respectively.　The　data　included　nine　correlated　factors　of　window　operation　in　three　wards　and　a　doctors＇　office.　The　results　indicated　that　the　hourly　window　opening　probability　of　the　wards　changed　regularly　with　time　from　10　to　65%,　whereas　the　probability　of　the　doctors＇　office　window　being　opened　remained　at　similar　to　70%,　which　was　quite　different　from　the　characteristics　of　other　building　types　in　Beijing　during　summer.　Among　the　correlated　factors,　the　outdoor　PM2.5　concentration　was　the　most　significant　for　two　room　types,　which　was　a　valuable　finding　for　window　status　analysis　in　hospitals.　Satisfied　prediction　accuracy　of　the　RF　model　was　obtained,　which　provide　a　reliable　reference　for　the　modeling　of　hospital　buildings.　This　research　could　deepen　the　understanding　of　the　behavior　in　hospital　buildings　and　provide　reference　for　future　modeling　research.