In　damage　assessment　of　composite　structures,　the　modal　curvature　appears　to　be　one　of　the　most　important　damage　indices　in　the　past　decades.　However,　a　noticeable　deficiency　of　the　modal　curvature　is　its　susceptibility　to　noise,　which　is　mainly　induced　by　the　numerical　difference　estimation.　This　study　proposes　the　scale-wavenumber　domain　filtering　method　based　on　the　combination　of　the　continuous　wavelets　transform,　the　discrete　Fourier　transform-based　modal　curvature　and　the　scale　wavenumber　domain　filtering　method.　The　continuous　wavelet　transform　provides　the　scale　domain　for　noisy　mode　shape　analysis,　in　which　the　normal　fluctuations　and　the　noise-induced　fluctuations　are　filtered　from　the　inspected　mode　shapes.　The　discrete　Fourier　transform-based　modal　curvature　supplies　the　wavenumber　domain　expressions　of　the　scaled　mode　shapes.　In　the　scale　and　wavenumber　domains,　some　special　filters　are　designed　and　used　in　noise　suppression.　The　effectiveness　of　the　proposed　method　is　analytically　verified　by　employing　the　cracked　composite　beam　model,　and　the　performance　is　further　validated　by　the　experimental　data　from　the　carbon-fiber-reinforced　polymer　beam　with　crack.　Based　on　these　validations,　it　is　observed　that　the　proposed　method　is　capable　of　revealing　slight　damage　in　noisy　condition,　without　the　requirement　for　the　prior　knowledge　of　material　properties.　(C)　2016　Elsevier　Ltd.　All　rights　reserved.