Asuperresolution(SR)methodforremotesensingimagesbasedonsparsedictionaryandstructuralself-similaritywasproposed.Furthermore,comparedwithcorrespondencebetweenLRandHRimagepatchesfromaconventionalSRmethods,twodictionarypairs,i.e.primitivesparsedictionarypairandresidualsparsedictionarypair,arerainingdatabase.Theprimitivesparsedictionarypairislearnedtoreconstructinitialhigh-resolution(HR)remotesensingimagefromasinglelow-resolution(LR)input.However,theinitialHRremotesensingimagelosessomedetailscomparewiththecorrespondingoriginalHRimagecompletely.Therefore,residualsparsedictionarypairislearnedtoreconstructresidualinformation.Finally,self-similaritystructuralwidelyexistinremotesensingimagesandthisfeaturecan beusedtocorrectthereconstructedimagebynonlocalmeans(NLM)method.Experimentalresultsshowedthattheproposedalgorithmprovidesbettersubjectiveandobjectivequality,whencomparedtotheconventionalalgorithmsanditsPSNRis24.6905,SSIMis0.7363