Expert judgements are potentially a valuable source of information in land degradation assessment, especially in areas where data paucity impedes the use of quantitative models. However, expert opinions are also much disputed because they are not tested for consistency, abstain from formal documentation, while their quantitative interpretation is inherently unidentifiable. This paper evaluates and formalizes the use of expert judgements to conduct a nationwide water erosion hazard assessment in Ethiopia. We, therefore, test the experts' judgement for consistency, its correlation with quantitative measurements on soil loss and its reproducibility. The study uses an Ethiopian and an international dataset for which experts gave qualitative judgements on water erosion hazard, for well-described sites under different types of land uses. Experts show a high consistency in their judgements on land degradation, but a comparison with quantitative soil losses from runoff plots reveals that boundaries of middle classes vary widely between experts, while they tend to overestimate soil loss. Reproducing expert opinions with an ordered logit model shows a reasonable accuracy in predicting presence or absence of erosion, yet the model is less precise in distinguishing between higher erosion classes. In 58 per cent of the cases, the model gives a similar classification as the experts, in 19 per cent the model estimates higher and, more seriously, in 23 per cent a lower erosion class. Mapping the model results for Ethiopia demonstrates a high erosion hazard for land under annual crop cultivation, while erosion under perennial crops, rangeland and forest is absent or moderate. The likelihood of the model to select the correct hazard class for rangeland is relatively high but medium to low probabilities prevail for erosion classes of other land uses. The model needs a further justification to give adequate results for arid areas. Copyright © 2003 John Wiley & Sons, Ltd.