The reactor pressure vessel (RPV) is a major component in all current nuclear reactor designs. It consists of several forged components joined through a welding process. The process of welding introduces residual stresses to the structure. Residual stresses are those which are present with no external loading on the structure. Plastic strains may also have been introduced to the heat affected zone (HAZ) of the weld. Both residual stresses and plastic strains are considered as load history effects. Cleavage fracture is a brittle mode of failure and would be catastrophic were it to occur in an RPV weld. The structural material used in the forgings that make up the RPV is ferritic steel, which, at the usual operational temperatures of nuclear plant, would not be susceptible to cleavage fracture. However, long-term neutron irradiation embrittles the RPV. This embrittlement, when combined with lower temperatures during routine plant shutdown events, renders cleavage fracture a concern. Most structural integrity assessments consider a single parameter fracture toughness value, obtained by testing deeply cracked specimens. It is known that geometrical factors such as reduced crack depths, or loading conditions such as tension instead of bending, increase the effective fracture toughness of the material, through a reduction in hydrostatic stresses, known as a loss of constraint. What is not clearly understood is the combined effects of the load history on constraint. The work presented in this thesis provides a methodology, using a 2- parameter fracture toughness approach, to predict failure through cleavage, at various levels of constraint, with both residual stresses and plastic strains. The methodology employs a failure curve in the J-Q space, derived from as received conditions, which is shown to well predict the toughness of specimens under the various initial conditions described above. This thesis presents an experimental programme and series of numerical analyses that enabled the derivation of the J-Q failure curve, and validated its ability to predict failure for various constraint levels and load histories.