This thesis describes the conceptualisation and development of the PYROCART model. This model simulates the spatial behaviour of fire in spatially heterogeneous environments. The principle aims of the research were to test the applicability of overseas fire spread models to New Zealand fuels, to investigate the environmental controls influencing wildland fire behaviour and to assess the applicability of Geographic Information Systems (GIS) to fire spread prediction. The PYROCART model integrates a Geographic Information System (Arc/Info) and the fire spread model of Rothermel ( 1972). The Rothermel model consists of a series of flux equations which describe the physical and chemical processes of combustion. Rate of spread is estimated to be the difference between these fluxes. A problematic limitation of this model is that it is assumed that the landscape in which the fire is being modelled is homogenous with respect to environmental descriptors such as fuel type, slope, wind speed and wind direction. The use of a cellular data model within a Geographic Information System overcomes some of the spatial limitations of the Rothermel model associated with the assumption of environmental homogeneity. The model is validated using a large wildfire which occurred on 27-28 May, 1995 on the west bank of the Cass River in the Cass Basin. This fire burnt 580 hectares across a complex vegetation mosaic comprising shrubland, stands of Nothofagus solandri var clif.fortioides, bog and tussockland. The pre-fire vegetation was mapped and fuel models were built for nine vegetation types. The topography and variation in the wind field of the fire scar were also surveyed. The overall prediction of the model is estimated to have an accuracy of 80%. Prediction accuracies within different fuel types, slopes and wind conditions are also presented and it is shown that fuel type and slope appear to be the dominant influence on fire spread. No trends in prediction accuracy by wind speed and wind direction are apparent. The predicted burned area and the real burned area have a similar overall shape. However, problems of over-prediction of backing and flanking rates of spread at high wind speeds are identified. The PYROCART model shows potential as a management tool, especially for the testing of hypotheses concerning alternative land management strategies. However, due to the complex input data and parameterisation techniques required to operate the model it is not suitable for in situ fire management.