In order to compare the influence weights of meshing stiffness and radial support stiffness on epicyclic gear train load sharing behavior, a translation-torsional dynamic model of compound planetary gear train was developed, including internal excitation such as stiffness, error and clearance. Taguchi method was used to establish a multi-factor analysis model of uniform load to calculate the uniform load coefficient of compound planetary gear train under the action of variable stiffness excitation. By using signal to noise ratio evaluation, analysis of variance and comparing the gradients for each stiffness excitation, the contribution ratios of each excitation were quantified and the stiffness excitation that determined load sharing behavior was found out. The research results indicated that different types of stiffness excitation had very different effects on load sharing. The radial support stiffness of the center wheel, the radial support stiffness of the planetary wheel and the meshing stiffness of all meshing pairs determined the load sharing behavior. Reducing the amplitude of stiffness excitation can effectively improve the load sharing behavior of the gear train.