On tilled fields runoff directions may be affected by tillage induced oriented roughness, causing runoff to Row along tillage lines instead of topographic direction. That this has an important effect on runoff and erosion patterns was already reported [Ludwig et al., Catena 25 (1995); Desmet and Govers, Catena 29 (1997); Souchere et al., J. Hydrol. 206 (1998); Takken et al., Catena 37 (1999)]. However, limited research has been carried out to develop models that can be used to predict whether flow will be in tillage or topographic direction. In this study a wide range of data was collected on runoff patterns observed in an agricultural catchment in the Belgian loess belt. The data show that for more than 75% of the mapped areas on hillslopes the dow was in direction of tillage. The data were analysed to develop two logistic regression models to predict runoff direction. The first model uses topographic slope, the angle between the tillage orientation and aspect direction and the degree of oriented roughness as input. In the second model, the effect of discharge on the Row direction is also considered using unit contributing area as a substitute variable. However, the application of the second model is complicated and error-prone. Furthermore, application of both models to a validation dataset showed only a minor increase in model performance when upslope area is included (95 vs. 93% of correct predictions). Therefore, it may be better to predict Row directions without taking discharge into account. The model without unit contributing area predicted very well the spatial variation of Row directions within a field surveyed by [Desmet and covers, Catena 29 (1997)]. Including this logistic model in runoff and erosion models will result in much better predictions of runoff and erosion patterns than can be obtained by using the traditional approach of calculating a runoff pattern based on topography only. (C) 2001 Elsevier Science B.V. All rights reserved.