Spatio-temporal progress of an epidemic of faba bean rust was monitored over a discontinuous field. Trap plots were sown at increasing distances from a source plot, in the centre of which plants were inoculated. Disease spread in the source plot followed a focal pattern, with a radial velocity of expansion slightly lower than 0.1 m per day. At the end of the experiment, all trap plots had been infected, and two of the most distant ones showed unexpected high disease severity. Using a three-dimension model of disease progress, we showed that the epidemics on the scales of the source plot and of the trap plots could not be combined into a single epidemic on the whole-field scale. The epidemics had equivalent infection rates on both scales, but changing the scale dramatically affected their distance parameter. The epidemic in the source plot could have been caused by a short-distance, high-frequency, deterministic mechanism of spore dispersal, whereas infection of the trap plots could have been caused by a long-distance, low-frequency stochastic mechanism of spore dispersal. Although our results agreed with the predictions of a simulation model postulating these two mechanisms, alternative hypotheses which could also explain the observed disease pattern remained to be tested.