The criteria for registering pesticides persistent in soil are still a matter of debate. Amongst modern pesticides, several triazole fungicides are very persistent, though no deleterious effects on soil microbial processes have been reported. The behaviour of ®ve such compounds (¯utriafol, epoxiconazole, propiconazole, triadimefon and triadimenol) has been examined in two ®eld trials utilising different agronomic treatments. These fungicides were applied in June 1996 at rates of 0.5 kg ha À1 , and soil cores were taken to 20 cm depth at intervals over 2.5 years and analysed by extraction and highpressure liquid chromatography. Triadimefon was quite rapidly reduced to triadimenol. Triadimenol, ¯utriafol and epoxiconazole were all very persistent with DT 50 b 400 days, whilst propiconazole had DT 50 c 200 days; behaviour was similar in the Rothamsted clay loam and Woburn sandy loam. Only ¯utriafol, the most polar and hence weakly sorbed of these fungicides, was appreciably leached, with traces reaching the 15±20 cm deep soil layer. Sprays applied to plots of fallow soil suffered loss of up to 50% of applied compound in the ®rst four weeks, a loss eliminated by shallow incorporation, indicating an early role for surface loss processes such as photolysis and/or volatilisation. A young barley crop intercepted about one-third of the spray, though subsequent rain caused some wash-off. After one to two years, amounts of each compound remaining in the plots were similar for the three agronomic treatments, especially for ¯utriafol, though with a tendency for the incorporated plots to have the most chemical and the barley plots the least. Computer simulation of behaviour in the ®eld using the model CALF, utilising sorption and degradation measurements made in laboratory incubations with these same soils together with daily climate measurements, overestimated persistence especially for ¯utriafol, epoxiconazole and triadimenol. This was due both to lack of inclusion of surface loss processes in the model, which caused initial deviations in the plots not receiving cultivation after spraying, and to a longer-term underestimation of breakdown in the ®eld. This latter was especially noticeable for triadimenol, which was not detected after 2.5 years despite predictions of c 50% remaining. Thus ®eld measurements of behaviour are desirable, as simulations based on laboratory measurements can overestimate persistence.