As part of a safety assessment of a component or structure, it is necessary to define in a rigorous manner the limits for its safe use and operation. This requirement leads to a need for accurate descriptions of the conditions for failure. In determining safe operating limits for failure by fracture, current methods are often overly pessimistic, especially following load history or in the presence of residual stress. Such conservatisms may lead to overdesign and excessive weight or premature removal of infrastructure from service. A study was conducted, described in this thesis, on the influences of previous load cycles on brittle fracture, primarily in A533B ferritic steel. Potential influences of remnant stresses on measured fracture toughness were explored by extracting test coupons from large scale welded components. Finite element simulations and experimental stress measurement were used to infer the effect on measured toughness. Re-analysis of previously published experimental data highlighted a range of limitations and practical problems with a number of current fracture criteria. To investigate the issues highlighted in greater depth, a program of fracture testing was conducted covering a wide range of specimen constraint levels and considering specimens with and without prior load history. The resulting fracture data set was used to study the applicability of numerous local approach methods, as well as crack tip fracture parameters, in terms of their transferability between geometries and ability to predict the effects of load history. It was shown that the effect of prior loading on fracture behaviour can be extremely significant. It was seen that the local approach, if properly calibrated, is able to predict the influence of load and geometry on fracture to an acceptable accuracy. It was also seen that consideration of fracture, even under brittle conditions, as a stress and strain controlled process improved the quality of the model predictions.