With　the　rapid　development　of　the　hyperspectral　technology,　spectra　unmixing　has　received　more　and　more　attentions.　In　this　paper,　in　order　to　measure　the　similarity　between　the　extracted　endmember　spectra　and　the　actual　corresponding　spectra　of　land　covers　and　maintain　the　spectral　absorption　feature,　a　wavelet　weighted　similarity　is　presented.　And　by　introducing　it　into　the　minimum　distance　constrained　nonnegative　matrix　factorization　method,　a　spectral　unmixing　method　based　on　the　wavelet　weighted　similarity　is　proposed.　Base　on　the　experiments　with　real　hyperspectral　image,　the　feasibility　and　real　performance　of　our　method　has　been　examined　and　compared　with　that　of　unsupervised　spectral　unmixing　method.