Soil desiccation cracks result in a decrease in irrigation efficiency and nutrients loss during irrigation events, and the prediction of crack preferential flow would provide theoretical basis for soil water management during crop planting. A dual-permeability model was proposed for crack preferential flow, based on the principle of water volume balance in which the sum of surface infiltration rate, laminar flow flux and lateral flow rate through crack walls equaled the surface irrigation intensity. The dual-permeability model was validated with help of dye tracer experiments of preferential flow. Furthermore, this preferential flow model was applied to practical use with rotation-combination design of initial/boundary conditions, i.e. antecedent water content and irrigation intensity. The results showed that the preferential flow patterns could be markedly classified into matrix flow (soil depth was 0~10cm) and preferential flow (soil depth was greater than 10cm). The variations of dye coverage were rapidly decreased when the soil depth exceeded 10cm. A comparison of the measured and simulated data showed that the dual-permeability preferential flow model was remarkably valid in predicting the characteristics of the surface infiltration rate, matrix flow depth and the dye coverage variations with depth. The Pearson correlation coefficient between the measured and the simulated data showed a significant level (P<0.01, R2 was 0.981). The propagation of crack preferential flow process could be divided into supply dominated phase, surface infiltration capacity/crack wall laminar flow dominated phase and surface/crack wall infiltration capacity dominated phase. Based on this model, a rotationcombination numerical simulation containing four levels of antecedent water content (θi was 0.20cm3/cm3, 0.25cm3/cm3, 0.30cm3/cm3 and 0.35cm3/cm3) and six levels of irrigation intensity (R0 was 0.10cm/min, 0.08cm/min, 0.06cm/min, 0.04cm/min, 0.03cm/min and 0.02cm/min) was conducted. The simulated preferential flow patterns were analyzed under 24 treatments. The results showed an increase in preferential flow fraction and a decrease in irrigation uniformity when the irrigation intensity was increased. As the antecedent water content was increased, the strength of preferential flow was decreased and the matrix flow depth became larger. In general, the proposed model was valid in predicting the propagating wetting fronts of crack preferential flow by adopting the water volume balance principle, which eliminated the disadvantages of the traditional dual-permeability preferential flow model that the roughness of the crack walls and the thickness of the laminar flow can not be accurately quantified. The proposed model provides a novel perspective into crack preferential flow prediction, and the simulation results would provide theoretical basis for the farmland water management and percolation loss.