Establishing　a　passenger　flow　prediction　mechanism　is　necessary　for　quickly　evacuating　many　passengers　in　an　emergency,　which　can　improve　the　service　quality　of　urban　rail　transit　(URT).　To　effectively　forecast　origin-destination　(OD)　passenger　flows　in　URT　under　emergency　conditions,　35-day　automatic　fare　collection　(AFC)　data　are　used　for　a　statistical　analysis　of　the　time,　location　and　passenger　flow　aspects.　The　influence　range　of　the　OD　passenger　flow　during　an　emergency　is　determined　by　analyzing　the　degree　of　passenger　flow　fluctuation.　Considering　the　time　period　of　an　emergency　occurrence　and　its　continuous　influence,　this　paper　also　studies　the　influence　of　an　emergency　occurring　at　a　station,　a　section　between　two　stations　or　a　section　across　several　stations.　A　spatial-temporal　correlation　prediction　model　of　OD　passenger　flow　based　on　nonlinear　regression　is　constructed　by　introducing　the　concept　of　passenger　flow　spatial-temporal　influencing　parameters.　According　to　the　characteristics　of　URT　lines,　a　passenger　flow　prediction　algorithm　is　proposed　to　predict　the　OD　passenger　flow　for　different　line　categories　for　an　emergency.　A　real　typical　emergency　involving　the　Beijing　urban　rail　transit　(BURT)　system　in　2017　is　analyzed　to　verify　the　effectiveness　of　the　proposed　model.　The　results　show　that　this　model　can　effectively　predict　OD　passenger　flow　in　a　URT　system　during　an　emergency,　which　provides　basic　support　for　the　evacuation　of　passengers.