Abstract: In order to study the influence of the net mean stress on the permeability coefficient of unsaturated remolded loess, an indirect method is used to predict the permeability coefficient through the generalized soil water characteristic curve which involves the effect of net mean stress on SWCC. To get the generalized soil water characteristic curve, four triaxial contraction tests are conducted for remolded loess samples by increasing matrix suction step by step and controlling the net mean stress as constant. Based on the measured generalized soil water characteristic curve, the Van Genuchten model for predicting the permeability coefficient of unsaturated soil is used. The permeability coefficient of every matrix suction under different net mean stresses is calculated. The results show that the permeability coefficient of unsaturated remolded loess decreases nonlinearly with the increase of the matrix suction in the semi-log coordinates. Under the same matrix suction, the smaller the net mean stress, the greater the permeability coefficient. The results may provide a useful basis for the deformation of high embankment and rainfall infiltration of slope stability evaluation in loess regions.