This paper describes an experimental and numerical study to assess the transposability of brittle fracture criteria from specimens of one type of geometry to another. The overall “master curve” approach, the Beremin model and a proposed model using the concept of critical stress were accordingly analysed. The experimental work supporting the analysis was made on 16MND5 reactor vessel steel. This was in the form of CT25 specimens, taken as the reference type, SENT specimens, ring specimens and CTpor specimens, which are CT specimens with a semi-elliptical surface defect. The analysis itself was made in two stages: the models were first calibrated on the basis of CT25 test results, then they were applied to specimens of other geometries. We then demonstrate that, in all cases, the models correctly replicated the variation of toughness (as measured on a CT25 specimen) with temperature. However, they all failed when applied to SENT and ring specimens, where calculation underestimates the probability of fracture. For CTpor specimens, the results are better, the master curve approach and the critical stress criterion give satisfactory results (but it has not yet been possible to apply the Beremin method). This paper concludes with a detailed analysis of the crack tip stress and strain fields, followed by an attempt to explain the differences between the different types of behaviour observed.