Abstract：The fluid flow of an actual infiltration experiment performed in fractured rocks is simulated with channel flow model. When building the flow model，a network of disc-shaped fractures is constructed with inverse method firstly. Then，after the removal of unconnected fractures(isolated fracture or fracture clusters not linked to the infinite cluster)，the fracture network is simplified to a network consisting of channels in three-dimensional space. When constructing the channel network，for each disc-shaped fracture in the network，nodes are designed at the centers of its intersections with other fractures and the center of the fracture-disc；and each node at intersection-centers is connected to the node at disc center by a channel. The diameters of the channels are modified with observed discharge data. This is a process to modify guessed input channel diameters over likely ranges of values until a satisfactory match between the simulated and observed discharges for the infiltration experiment is obtained. It is impractical to modify the diameter of each individual channel. Consequently，it is assumed that，within a given fracture，the diameters of channels follow the lognormal distribution，so，only the mean and standard deviation of the channel diameters for each fracture set are corrected. To validate the stability of the model，20 fracture networks are generated and the flow in them are investigated. The difference between the measured and the average discharges of the 20 realizations is about 6%. That is，if the discharge of the experiment is predicted with 20 realizations of the stochastic model，the error of the prediction is about 6%.
