Abstract: The large-well-method is one of the common methods in the prediction of mine water inflow, but sometimes the errors occur in the prediction, the mainly reason is that the aquifer thickness and the permeability coefficient can't reflect the actual situation and the man-made influence is big. Through statistical analysis of the regional structural complexity and the permeability coefficient obtained from pumping test, the positive relationship was found between the structural complexity and permeability coefficient, by using the fractal theory, the value of permeability coefficient is optimized for large-well-method to be closer to the real situation. Aiming at the changes of water inflow in different aquifers above the roof and below the floor influenced by mining disturbance, the aquifer thickness was weighted cumulatively, the water inflow was the sum of different aquifers. After parameter optimization, the arbitrariness of the parameters selection was avoided, applying the improved large-well-method at working face 2190 in Fangezhuang coal mine, the precision of prediction has been greatly improved.