Abstract: Grain grade and void ratio are the crucial factors affecting the coefficient of permeability for coarse-grained soil. In this paper, 93 sample data of full grain grades (d₁₀~d₁₀₀) and void ratio are collected to analyze and predict the coefficient of permeability by applying the BP neural network optimized by genetic algorithm. By means of the mean impact value (MIV) method and verification test, the influential extents of each grain size and void ratio on the coefficient of permeability are evaluated and discussed. The results reveal thatd₅₀is the critical particle size of coarse-grained soil, indicating that if the others grain sizes keep constant, the coefficient of permeability will increase with the increasing grain size belowd₅₀and decreasing grain sizes overd₅₀. The influential extent of the fine particles smaller thand₅₀is much more than these coarse particles bigger thand₅₀. According to the relative influential weights, d₂₀, d₈₀, d₄₀belong to the high-sensitivity grain sizes, d₁₀, d₅₀, d₁₀₀, andd₇₀are medium-sensitivity grain sizes, andd₃₀, d₉₀, andd₆₀are low-sensitivity grain sizes. Moreover, the effect of void ratio on the coefficient of permeability is greater than that of each grain size, and the coefficient of permeability is positively correlated with the void ratio. As for the coarse-grained soils with the same grain size, the variation in void ratio will result in the change in the coefficient of permeability by order of magnitude. It is concluded that the coefficient of permeability of coarse-grained soil can be well predicted by applying the GA-BP Neural Network taking full grain grades and void ratio into account.