Population viability analysis (PVA) packages do not always allow realistic simulation of particular life cycle features, so they may produce unrealistic predictions of extinction risk. This was suspected for a cycling Soay sheep population, Ovis aries L., that grows, overeats its habitat and then suffers high winter mortality. We compared projections of PVA models for the sheep that incorporated either an unrealistic ceiling carrying capacity using INMAT (the only choice) and VORTEX (the default), or realistic density dependence for survival (VORTEX). At year 50, the ceiling models predicted extinction probabilities of 60.4% (INMAT) and 87.4% (VORTEX), compared to only 4.6% for the density-dependent model. Small populations were equally likely to increase or decrease with the ceiling models, while they had high probabilities of increase with the density-dependent model, as found in reality. PVA cannot be relied upon to produce realistic projections if inappropriate mechanisms of population regulation are used. (C) 2000 Elsevier Science Ltd.