In European loess belt soils, infiltration and erosion processes are strongly influenced by surface crusting. Modelling infiltration into these crusts has led to the development of equations of varying complexity, ranging from simple empirical equations to numerical solution of the Richard's equation. However, a number of issues important for modelling effective erosion at the catchment scale remain unsolved. The objective of this study was to contribute to the elaboration of an expert-based runoff prediction model able to simulate the influence of soil conservation practices in the context of loess soils susceptible to crusting. Experiments have been implemented both in the laboratory and in the field at various scales ranging from small plots up to catchments. The experimental results provided a set of reference infiltration and runoff data under a variety of different situations in terms of weather conditions, surface state, land use and agricultural practices. Infiltrability ranged from 2 mm h−1 for crusted surfaces up to more than 30 mm h−1 for undegraded surfaces. These references were used to develop decision rules in the forms of matching tables to characterise agricultural fields with an infiltration capacity for a given rainfall event. For the area of the Pays-de-Caux (Normandie, France), we defined five potential infiltrability classes of 2, 5, 10, 20 and 50 mm h−1. A runoff circulation network calculated from a Digital Elevation Model (DEM) combined with information on field operations allows the calculation of total runoff volume for a rainfall event at any point of the catchment. Calculated and measured runoff for a first series of events were in satisfactory accordance.