A round-robin investigation has been performed, where stress analysts from eight different organisations carried out a total of 11 predictions of the expected fatigue limit of a diametrically loaded cast ring subjected to fluctuating tensile or compressive loading. Whereas geometry, load parameters, and type and quality of material (spheroidal graphite cast iron EN-GJS-600-3) had been prescribed, the participants were free to use computational tools and models, and fatigue assessment models and data of their own choice. The objectives of the investigation were to compare the 11 predictions (i) among themselves, and (ii) with a posteriori experimental fatigue limits determined by means of stair-case testing. The fatigue limit predictions showed coefficients of variation of as large as 25%. Even for a group of analysts from a single organisation, the coefficients variation were around 15%. Fatigue tests gave mean fatigue limits 60% (tensile loading) and 30% (compressive loading) above the a priori predictions. Possible reasons for the large deviations between single predictions and for their conservatism have been proposed. It seems that design engineers (i) make use of the available room for interpretation of models and data, and (ii) have an unconscious tendency to make conservative assumptions. Only if models and data for fatigue assessment are prescribed in great detail, can the ‘scatter’ among fatigue limit predictions be expected to decrease below 15–25%. Improved ‘absolute’ predictions would require more accurate fatigue data.