Diffuse reflectance spectroscopy is a useful tool for obtaining quantitative information in turbid media, which is always achieved by developing a multivariate regression model that links the spectral signal to the component concentrations. However, in most cases, variations between the actual measurement and the modeling process of the device may cause errors in predicting a component's concentration. In this paper, we propose a data-processing method to resist these variations. The method involves performing a curve fitting of the multiple-position diffuse reflectance spectral data. One of the parameters in the fitting function was found to be insensitive to inter-device variations and sensitive to the component concentrations. The parameter of the fitted equation was used in the modeling instead of directly using the spectral signal. Experiments demonstrate the feasibility of the proposed method and its resistance to errors induced by inter-device variations.