Measurement　errors　occur　in　many　real　data　applications.　In　this　paper,　the　linear　and　the　non　linear　wavelet　estimators　of　the　derivatives　of　the　density　function　are　constructed　in　the　case　of　data　contaminated　with　heteroscedastic　measurement　errors.　We　establish　L-p　risk　performance　of　the　estimators　and　show　that　they　achieve　fast　convergence　rates　under　quite　general　conditions.