In the present paper, a new method named response variable transform (RVT) is proposed to reduce the effect of scattering in the NIR spectroscopy measurement. In the RVT method, the rates of the change in light intensity at different spatial positions with the concentration of the investigated composition are firstly sampled. Then, based on the sampled changing rates of light intensity, the absorption coefficients, which are used as new response variable, are calculated using the inverse adding-double (IAD) algorithm. Finally, through the partial least squares (PLS) regression, the calibration models for the investigated compositions are established using these new response variables. To validate the feasibility and effectiveness of the proposed method, the RVT method is applied to reduce the effect of scattering during the process of composition concentration measurement in milk. First, the relationships between the absorption coefficient and the concentrations of major compositions including fat and protein are analyzed. Then, with Monte Carlo simulations for different concentration levels of fat and protein, the influence of scattering on the relationship between the absorbency and the concentration is investigated and verified. Finally, for comparison, both the RVT method and the traditional preprocessing techniques are used to reduce the effect of scattering during the process of prediction model establishment. Experimental results showed that the prediction precision for fat and protein with the RVT method is increased by up to 33.9% and 46.5% than that with multiplicative signal correction (MSC) algorithm, and the prediction precision for fat and protein with the RVT method is increased by up to 55.9% and 33.7% than that with standard normal variate (SNV) algorithm, which also indicates that the RVT method is an effective method for eliminating the interference of scattering in the NIR spectroscopy measurement.