In　order　to　diagnose　the　root　causes　of　stampede　accident　in　scenic　spots　under　uncertainty,　this　paper　proposes　an　intuitionistic　fuzzy　Bayesian　network　bidrection　reasoning　model.　First,　we　use　multi-granularity　trapezoidal　intuitionistic　fuzzy　sets　to　express　experts＇　judgements　about　fault　possibilities　of　root　nodes,　and　extend　the　fuzzy　possibility-probability　transformation　function　to　obtain　the　failure　prior　probabilities　of　root　nodes.　Second,　we　use　the　D-S　evidence　theory　to　conduct　information　fusion　and　obtain　the　aggregated　prior　probability　values　of　root　nodes.　We　perform　Bayesian　network　bidirection　reasoning　to　conduct　Bayesian　network　updating　analysis　and　important　analysis.　And　then,　the　proposed　approach　is　implemented　in　the　empirical　study　in　the　Mount　Hua　scenic　spot,　China.　The　proposed　method　can　overcome　the　information　loss　caused　by　the　previous　defuzzication　method　in　estimating　the　prior　probabilities　of　root　nodes,　and　has　advantages　in　solving　the　problem　of　fault　diagnosis　and　Bayesian　inference　under　uncertainty.　©　2022　Systems　Engineering　Society　of　China.　All　rights　reserved.