This　study　develops　a　theoretical　model　to　estimate　the　scaling　error　of　the　two-band　ratio　of　red　to　near-infrared　(TBRRN)　in　an　inhomogeneous　pixel.　Three　different　imageries　and　a　3　x　3　moving　window　are　used　to　verify　and　approximately　estimate　the　scaling　error　of　the　TBRRN　for　remote-sensing　imageries.　The　datasets　are　Landsat　Thematic　Mapper　(Landsat/TM)　imagery　taken　on　15　October　2005,　satellite　probatoire　d＇Observation　de　la　terre　(SPOT)　imagery　taken　on　7　September　2005,　and　moderate-resolution　imaging　spectroradiometer　(MODIS)　imagery　taken　on　5　October　2005　of　the　Yellow　River　Estuary.　It　is　found　that　1)　about　15.70%,　17.24%　and　26.52%　of　SPOT,　Landsat/TM　and　MODIS　pixels　have　relative　scaling　error　higher　than　2%　respectively,　2)　the　average　relative　scaling　error　increases　with　increasing　scale　of　the　image　pixel,　and　3)　it　is　difficult　to　achieve　the　goal　of　the　National　Aeronautics　and　Space　Administration　of　obtaining　valid　ocean-color　data　of　the　world＇s　oceans　for　estimating　the　chlorophyll-a　concentration　with　uncertainty　of　less　than　35%　if　the　scaling　error　cannot　be　effectively　reduced.　Our　results　suggest　the　need　for　an　in-depth　study　of　scaling　errors　in　water-color　remote　sensing.