There are usually three calculation methods of normalized water-leaving radiance in use of in-water radiometric profile data the local linear regression method is the commonly used one for its simple principle and processing. However this method need manual selection of the regression depth interval the results are mostly affected by subjective factors of processors which can lead to 5%-45% errors with different depth intervals selected. In this paper a statistical method is introduced that calculates all the results of linear regression in the sub-interval of selected depth interval removing the exceptional values of them and the mean of remained values is best result. The other advantage of this method is that we can find the best regression depth intervals in use of the result. At last the author compares the results of processes of three measurements: in-water radiometric profile measurement above-water radiance measurement and the reversion produce of SeaWiFS data at the same time and site the results show that this statistical method is available and looking Es(O<sup>+</sup>.λ) on as the validation of extrapolation of Ed(O<sup>-</sup>,λ) is incorrect.