Artificial　neural　network　(ANN),　the　evidential　reasoning　(ER)　approach　and　multiple　regression　analysis　(MRA)　can　all　be　utilized　to　model　bridge　risks,　but　their　modelling　mechanisms　and　performances　are　quite　different　and　therefore　need　comparison.　This　study　compares　the　modelling　mechanisms　of　the　three　alternative　approaches　and　their　performances　in　modelling　a　set　of　bridge　risk　data.　It　is　found　that　ANN　outperforms　the　ER　approach　and　MRA　for　the　considered　case　study.　The　reason　for　this　is　analyzed.　The　advantages　and　disadvantages　of　the　three　alternative　approaches　are　also　compared.　©　2005　Elsevier　Ltd.　All　rights　reserved.