In the boreal forest of North America, as in any fire-prone biome, three environmental factors must coincide for a wildfire to occur: an ignition source, flammable vegetation, and weather that is conducive to fire. Despite recent advances, the relative importance of these factors remains the subject of some debate. The aim of this study was to develop models that identify the environmental controls on spatial patterns in area burned for the period 1980-2005 at several spatial scales in the Canadian boreal forest. Boosted regression tree models were built to relate high-resolution data for area burned to an array of explanatory variables describing ignitions, vegetation, and long-term patterns in fire-conducive weather (i.e., fire climate) at four spatial scales (10(2) km2, 10(3) km2, 10(4) km2, and 10(5) km2). We evaluated the relative contributions of these controls on area burned, as well as their functional relationships, across spatial scales. We also assessed geographic patterns of the influence of wildfire controls. The results indicated that extreme temperature during the fire season was a top control at all spatial scales, followed closely by a wind-driven index of ease of fire spread. However, the contributions of some variables differed substantially among the spatial scales, as did their relationship to area burned. In fact, for some key variables the polarity of relationships was inverted from the finest to the broadest spatial scale. It was difficult to unequivocally attribute values of relative importance to the variables chosen to represent ignitions, vegetation, and climate, as the interdependence of these factors precluded clear partitioning. Furthermore, the influence of a variable on patterns of area burned often changed enormously across the biome, which supports the idea that fire-environment relationships in the boreal forest are complex and nonstationary.