The　problem　of　estimating　a　function＇s　derivative　arises　in　many　scientific　settings,　from　medical　imaging　to　astronomy.　In　this　paper,　we　consider　a　hard　thresholding　wavelet　approach　to　certain　blind　deconvolution　problem　based　on　the　heteroscedastic　data.　Motivated　by　Delaigle　&　Meister＇s　work,　the　adaptive　wavelet　estimators　are　proposed　and　their　asymptotic　properties　are　investigated.　It　is　shown　that　the　derivative　estimators　for　the　blind　deconvolution　model　are　spatially　adaptive　and　attain　the　optimal　rate　of　convergence　up　to　a　logarithmic　factor　over　a　range　of　Besov　classes.　Our　theorems　generalize　the　results　of　Cai　and　Navarro　et　al.　in　some　sense.