We have investigated the effects of domain and grain anisotropy on spin-resonance in magnetic permeability, implementing a Monte-Carlo approach and a coupled Landau-Lifshitz-Gilbert equation. The Monte-Carlo approach provides great flexibility by employing different probability density functions, allowing modeling of material texture differences that may occur due to different preparation methods. Changes in the permeability tensor result from variations in grain demagnetization and domain demagnetization as well as the anisotropy field relative to saturation magnetization. Experimental permeability measurements on demagnetized polycrystalline yttrium iron garnet show for the first time that the best fit to measured data requires a complex distribution of both grain and domain demagnetization factors. Assuming that grain and domain demagnetizations are decoupled, it was found that the grain structure (i.e., grain demagnetization distribution) has a smaller effect on the frequency-dependent permeability than does the same distribution of domains (i.e., domain demagnetization distribution). Implications for modeling experimental data assuming particular phenomenological loss coefficients or linewidths are also offered.