Across western North America, fire regimes (i.e., the frequency, extent, and severity of fire events) are changing in response to warming climate. Regions in which fire regimes are driven by top-down controls (e.g., climate, fire weather) are likely to see the largest future increases in area burned as climate continues to warm. These climate-limited fire regimes are characterized by infrequent, typically high-severity stand-replacing fire with multi-century fire return intervals and fire events as large as one million hectares in size. Managers working in regions with climate-limited fire regimes are preparing for a warmer, drier future that may bring more frequent, extreme fire events. One need highlighted by managers and punctuated by the widespread west coast fire activity in summer 2020 is an understanding of the historical patch structure of wildfires, particularly the total area burned and high-severity patch size distribution of individual fire events and their range of variation. However, due to the historically infrequent nature of fire in climate-limited fire regimes, information about the size and severity of fire events in these regions is inherently limited. Using a satellite fire severity dataset of 1,615 fire events occurring across the Northwest US between 1985 and 2020, we present an approach for characterizing burn severity patterns expected within contemporary climate-limited fire regimes. We asked: (Q1). What is the relationship between overall fire size and high-severity patch structure of contemporary fires (1985-2020) in the Northwest US? (Q2). Does this relationship vary by region, time period, or across a gradient of fuel- to climate-limited fire regimes? To address Q1, we use nonparametric quantile regression to quantify the range of variation in high-severity burn patch metrics expected across the range of observed fire sizes (400 – >400,000 ha). To address Q2, we test whether the relationships between patch metrics and fire size vary by geographic region (Pacific Northwest versus Northern Rockies), year, time period (1985 – 2000 versus 2001 – 2020), or fire regime (infrequent and high-severity versus moderately frequent and mixed-severity or frequent and low-severity). We observed: 1) high-severity patches increasing consistently in size and spatial homogeneity with greater fire size within fire regimes, 2) different ranges of variation in scaling relationships among fire regimes, and 3) widely varying distributions of high-severity patches within smaller fires but convergence of patch-size distributions toward a power law function with increasing fire size. Collectively, our results suggest spatial patterns of high-severity fire demonstrate clear and consistent scaling behavior. Within fire regimes, scaling relationships did not differ substantially across space or time, suggesting that as fire size distributions potentially shift under climate change, the stationarity we observed in patch-size scaling can be used to infer expected future patterns of burn severity. Managing for future fire requires not only projecting possible changes in regional metrics such as annual area burned, but also anticipating the potential ecological impacts of those changes. At broad scales, stationarity in scaling relationships offers a means of projecting the potential range of ecological impacts expected with future fire activity. Continued implementation of the methods presented here would permit changes in scaling relationships to be detected (e.g., downward shifts in scaling relationships might suggest an increasing prevalence of local-scale fuel constraints) that might signal important future changes in the nature of fire regimes.