Modeling preferential flow in soils is still a challenge for the scientific community working on water resources. Indeed, it is an issue to determine the functional parameters of models dedicated to water flow, that are currently obtained by fitting processes, whereas their relationships with soil structure remain poorly known. Improved models are expected from a better understanding of the links between functional and structural parameters, which can be achieved thanks to recent developments in imaging methods such as X-ray Computed Tomography (CT). The paper seeks to improve a dual-porosity model, coupling matrix flow (by Richards equation) and preferential flow (by a Kinematic Dispersive Wave), by substituting some model parameters, usually obtained by inversion of experimental data, by those assessed from CT images of the soil structure. Thus, two versions of the model are compared, the "classical" and the "advanced" one including parameters determined using the 3D images of the sample structures. To compare model versions with real situations, infiltration experiments were conducted in lab on two different soils at two initial water contents. An X-ray medical scanner allowing acquisitions of large soil volumes (approximate to 1700 cm(3)) with a voxel size of 400 pm was used to image the sample structures. Then, we derived two geometrical parameters from the macroporosity network: the percolating macroporosity and a characteristic dimension of this macropore network, the mean inter-macropore distance. The sensitivity analysis conducted on the classical version of the model showed that the kinematic coefficient and the dimensional parameter of the porous medium are the two main contributors to the cumulated drainage whatever the initial condition. Although experimental data are better simulated by the classical version of the model, drainage dynamics is also well simulated by the advanced version. However, differences between the model versions that are small for both soils at field capacity become significant for the dried state (mean initial matric potential of -3.5 m). This emphasizes the crucial effect of the sink-source term and in particular the complex effect of the dimensional parameter that it contains. Indeed, difficulties to simulate properly water exchange between porosity domains are encountered for both versions of the model. We conclude that empirical parameters that were up to now fitted from experiments could be deduced from geometrical indicators computed from CT images and that owning to these first results the applied methodology is promising to achieve a better understanding and modeling of preferential flow processes and to improve model predictability.