A survey was set up in Central Belgium in order to assess rill erosion rates on arable land. Mean rill erosion rate for all fields in the survey during a three year period was 0.36 kg/m2. Statistical analysis of the data allowed to calculate mean estimates for the slope and length exponent of 1.45 and 0.75 respectively. These results agree well with other data. Mean values should be used with caution as a rather important variation in slope and length exponents was found which could not be explained. Apart from topography, vegetation cover is by far the most important control on between-field variations in rill erosion rate. Other significant factors are soil texture, aggregate stability and, for silty soils, the state of the surface layer. Generally, the available information allows only a limited explanation of the observed variation in rill erosion rates. Although it is believed that further improvement is certainly possible, a considerable margin of error on individual predictions has to be accepted, not only because the process mechanics of rill erosion is not very well known, but also because the controlling factors cannot all be measured within the framework of an extensive survey. It may therefore be useful to express predictions in probabilistic terms.