AbstractMelt percolation through partially molten “mushy” regions of the crust underpins models for magma migration, accumulation, and processes that prime systems for eruption. Knowledge of the hydraulic properties of magma mush, specifically permeability, is required for accurate predictions of melt migration rates and accumulation timescales. Previous studies, validated for cuboidal crystal analogs, show that crystal shape exerts a first‐order control on the permeability, and is tested here for anisometric natural crystal shapes using X‐ray CT 3D data sets of magma mush analogs made from packed confectionary particles arranged randomly. We use a lattice‐Boltzmann fluid flow simulation tool to determine the permeability of the analogue melt phase network between the packed particles. We find excellent agreement with our data sets to within ∼0.1 log units, when the specific surface area is measured. To extend this to more typical cases where the specific surface area is unknown, we use the shape and size of the objects determined in both 3D and 2D to estimate the specific surface area assuming a cuboid approximation. These approximate solutions give good results to within ∼0.5 log units of the measured permeability and offer a method by which permeability could be estimated from a thin section of a cumulate or pluton sample. Our shape‐sensitive approach is more accurate than existing models for permeability of magma mush, most approximating natural crystal shapes to spheres. We therefore propose that these could be implemented in dynamic magma mush models for melt movement in the crust to produce more accurate flux predictions.