The motivation for the research described in this thesis was to develop a simple method for dealing with one specific, but fundamentally important, aspect of the variability that is inherent in modelling fires in buildings; namely, the variability in the heat release rate (HRR) curve that is used as one of the inputs for performance-based fire safety engineering calculations. The traditional approach by fire safety engineers is to select a single (or a few) HRR curves, and conduct a deterministic analysis of the fire environment. In doing so, no attempt is generally made to investigate the range of possible HRR curves that could occur in the particular occupancy of interest. The development of a new method is therefore presented in this thesis which offers a rational way to quantify the variability in pre-flashover HRR curves for residential-scale occupancies – the methodology is termed a ‘design fire generator’, or DFG. The DFG is in turn a module within a new quantitative risk analysis fire model called B-RISK. The research described in this thesis forms part of a larger, joint research project undertaken by BRANZ and the University of Canterbury, with the primary output from the overall project being the B-RISK computer model. The approach taken by B-RISK is to use Monte-Carlo simulation for an iterative series of deterministic fire modelling calculations. As well as sampling from various probability distributions for different input parameters, B-RISK also requires a unique HRR curve, for each iteration, and hence the need for the DFG. The conceptual basis for the DFG is random population of a compartment with combustible items, a first item randomly igniting and then fire spread to secondary items occurs, based on incident radiation and an associated ignition criterion. As secondary items ignite the principle of superposition is used to generate a composite HRR curve for the compartment of fire origin, which becomes an input to one of multiple Monte-Carlo iterations in B-RISK. It is demonstrated that the DFG can successfully quantify the variability in pre-flashover HRR curves, the primary aim of the project. At the same time, the DFG was more conservative than the corresponding HRR curve required by the Verification Method, C/VM2, for this type of occupancy. As well as making theoretical predictions of HRR curves, the DFG can be used to recreate specific compartment fire scenarios. A series of blind modelling and compartment fire experiments were also conducted to benchmark the predictive capability of the DFG/B-RISK in this regard. The trueness of the predictions of the ignition of the first secondary item were in good agreement with the compartment experiments, but thereafter the DFG predicted ignition times that were generally earlier than occurred in the experiments, resulting in a peak HRR being predicted to occur earlier than the experiments. Further analysis of the data also indicated that the DFG not allowing for compartment effects to enhance the HRR of secondary items also contributed to the low trueness of the overall HRR predictions.