In spectrochemical quantitative analysis of solutions containing scattering components, the spectral nonlinearity caused by scattering seriously affects the prediction accuracy, robustness, and even feasibility of the models. Unlike the traditional methods (modeling with the spectra data of single pathlength) of approximating the nonlinear spectral line to linear to reduce the nonlinear features of scattering, a new method is proposed to reduce the effect of scattering by taking advantage of the nonlinear characteristics of spectral lines. First, the logarithmic function is used to fit the attenuation of multiple pathlengths, then the regression coefficient of the function is taken as the characteristic parameter of scattering, and the wavelengths with smaller characteristic parameter are selected as the modeling wavelengths. The model is robust and insensitive to the effect of scattering. The experiment involving a variety of scattering cases containing intralipids and ink was taken to verify the method. An F-test of the experimental results was significant at the 0.05 level. The root mean square error of prediction of the new method was 1.94%, and the prediction accuracy was 75.5% higher than that of the traditional model. The new method provides a novel approach toward describing the spectral nonlinearity with a function. [ABSTRACT FROM AUTHOR]