To control and use wildland fires safely and effectively depends on creditable assessments of fire potential, including the propensity for crowning in conifer forests. Simulation studies that use certain fire modelling systems (i.e. NEXUS, FlamMap, FARSITE, FFE-FVS (Fire and Fuels Extension to the Forest Vegetation Simulator), Fuel Management Analyst (FMAPlus), BehavePlus) based on separate implementations or direct integration of Rothermel's surface and crown rate of fire spread models with Van Wagner's crown fire transition and propagation models are shown to have a significant under prediction bias when used in assessing potential crown fire behaviour in conifer forests of western North America. The principal sources of this under prediction bias are shown to include: (i) incompatible model linkages; (ii) use of surface and crown fire rate of spread models that have an inherent under prediction bias; and (iii) reduction in crown fire rate of spread based on the use of unsubstantiated crown fraction burned functions. The use of uncalibrated custom fuel models to represent surface fuelbeds is a fourth potential source of bias. These sources are described and documented in detail based on comparisons with experimental fire and wildfire observations and on separate analyses of model components. The manner in which the two primary canopy fuel inputs influencing crown fire initiation (i.e. foliar moisture content and canopy base height) is handled in these simulation studies and the meaning of Scott and Reinhardt's two crown fire hazard indices are also critically examined.