A three-flux approximation of the equation of radiative transfer is used to separately determine the effective specific scattering and absorption coefficients of powder mixtures from hemispherical reflectance and transmittance measurements. For a two-component mixture of lactose and paracetamol, it is demonstrated how the knowledge of the separately known scattering coefficient can be used to improve partial least-squares regression (PLS) calibrations of diffuse reflectance data pretreated by multiplicative scatter correction (MSC). Furthermore it is shown that the measured specific absorption coefficient of the investigated mixtures is not generally a linear function of the constituents concentration, a result which might be caused by the mixing procedure of the samples. With the use of the absorption coefficient, it is demonstrated that artificial neural networks are superior to PLS calibrations when modeling a nonlinear relation.