Such　complex　causative　factors　in　current　failure　probability　models　are　represented　by　simply　random　uncertainty　and　completely　independent　or　correlation　of　failure　modes,　which　can　often　limit　the　model　utility.　In　this　study,　we　developed　a　methodology　to　construct　failure　probability　models　for　high　fill　levees,　incorporating　the　identification　of　uncertainties　and　an　analysis　of　failure　modes.　Based　on　quantification　of　stochastic-grey-fuzzy　uncertainties,　probability　analysis　involved　with　overtopping,　instability　and　seepage　failure　modes　was　implemented　combined　with　probability　and　non-probability　methods.　Given　that　the　interaction　among　failure　modes　typically　exhibits　nonlinear　behavior,　rather　than　linear　correlation　or　complete　independence,　a　simple　methodology　for　the　binary　Copula　function　was　established　and　implemented　in　MATLAB.　This　methodology　was　applied　to　the　high　fill　segments　of　a　long-distance　water　transfer　project　characterized　by　high　population　density.　It　shows　that　the　failure　probability　of　a　single　failure　mode　is　overestimated　when　uncertainties　are　not　considered,　because　of　the　randomness　and　fuzziness　of　some　parameters　and　the　greyness　of　information.　Meanwhile,　it　is　found　that　the　magnitude　of　failure　probability　related　to　levee　breach　is　overestimated　without　respect　to　failure　modes　correlation,　especially　when　the　probabilities　of　seepage　and　instability　are　both　significant　and　closely　aligned.