In wildland–urban interface areas, firefighters balance wildfire suppression and structure protection. These tasks are often performed under resource limitations, especially when many structures are at risk. To address this problem, wildland firefighters employ a process called “structure triage” to prioritize structure protection based on perceived defensibility. Using a dataset containing triage assessments of thousands of structures within the Western US, we developed a machine learning model that can improve the understanding of factors contributing to assessed structure defensibility. Our random forest models utilized variables collected by wildland firefighters, including structural characteristics and the surrounding ignition zone. The models also used landscape variables not contained within the triage dataset that captured important information about accessibility, vegetation, topography, and structure density. We achieved a high overall accuracy (77.8%) in classifying structures as defensible or non-defensible. The presence of a safety zone was the most important factor in determining structure defensibility. Road proximity, vegetation composition, and topography were also found to have high importance. In addition to improving the understanding of factors considered by wildland firefighters, communities could also gain from this information by enhancing their wildfire response plans, focusing on targeted mitigation, and improving their overall preparedness.