Models that predict the distributions and habitat suitability for species of conservation concern can be useful for guiding survey, monitoring, and conservation planning efforts. The Humboldt marten (Martes caurina humboldtensis) has significantly declined throughout its historical range in coastal California and Oregon and this taxon is known from a few remnant populations. We developed a landscape habitat suitability model to identify areas of suitable habitat not yet surveyed, to provide a template for designing monitoring programs and research studies, and to inform the development of a conservation strategy. We used the results of 1,159 occupancy surveys to develop a predictive habitat model using Generalized Additive Modeling to relate Humboldt marten detections to combinations of environmental and habitat attributes hypothesized to influence marten distribution. We measured 30 candidate variables at three spatial scales, using 0.5, 1.0, and 3.0 km, and although several models were competitive, each contained the same three core variables: old growth structural index measured at the 1-km scale, serpentine habitat measured at the 3-km scale, and annual precipitation measured at the 3-km scale. The final model had an overall correct classification rate of 91% for marten detections, 82% for non-detections and a true skill statistic of 0.73. Model predictions were stable when cross-validated, with the correct classification of marten detections (89%) varying little. The largest complexes of predicted suitable habitat occurred in the areas with the three extant marten populations (north coastal California, south coastal Oregon, and central coastal Oregon), however connectivity of suitable habitat to areas outside these three areas appears limited.