The core of the satellite characteristics inversion based on mixed satellite spectra is the mathematical model and inversion algorithm. Theoretical model of spectral mixing was built with experiments conducted to justify the model. First, theoretical analysis of components′ spectral scattering model, linear spectral mixing model and unmixing methods of satellite′s spectral data was conducted. Then, experiments were designed to measure and calibrate the spectral BRDF of a high-fidelity GEO satellite and its components, while the spectral scattering characteristics of component and material were discussed. Finally, nonnegative constrained least square methods were utilized to unmix the satellite′s spectral data, with the largest relative residue less than 10%. Experiment results show that the linear spectral mixing model and nonnegative constrained least square unmixing methods have practical meaning in explaining spectral data of satellites and inversing satellite conditions.