Wildfires shape the distribution and structure of vegetation across the inland northwestern United States. However, fire activity is expected to increase given the current rate of climate change, with uncertain outcomes. A fire impact that has not been widely addressed is the development of unburned islands; areas within the fire perimeter that do not burn. These areas function as critical ecological refugia for biota during or following wildfires, but they have been largely ignored in methodological studies of remote sensing assessing fire severity under the assumption that they will be detected by algorithms for delineating fire perimeters. Our objective was to develop a model for classifying unburned areas within wildfire perimeters using moderate resolution satellite (i.e., Landsat) and ancillary data. We performed field observations at locations that were unburned or lightly burned within the perimeters of 12 wildfires that burned in 2012 and 2014, and augmented this with field data previously acquired on another seven wildfires across the study region. We used randomForest and classification trees to separate burned from unburned locations with high overall classification accuracy (91.7% and 89.2%, for randomForest and classification tree methods respectively). Classification accuracy was significantly higher than the semi-automated classification products from the Monitoring Trends in Burn Severity (MTBS) program. After application of the most parsimonious and accurate classification tree model, we found that the average unburned proportion of the fires was 20% with high variability between fires (standard deviation: 16.4%). The total area of unburned islands in non-forested areas was significantly higher than the total unburned area in forested areas. Accurate detection and delineation of unburned areas is increasingly critical, as some of these unburned areas contain habitat (i.e., wildfire refugia) that are crucial for maintaining biodiversity and functioning of ecosystems, particularly given observed and projected anthropogenic climate change.