Soil inventory information is sparse to non-existent in remote and rugged areas of Washington State. Effective management of these remote areas requires a baseline of information for the variety of natural resources present, including soil type and pedologic conditions. In order to provide the first generation of soil survey information for remote areas, a GIS-based model of soil distribution was developed through graduate research at Washington State University. Our Remote Area Soil Proxy (RASP) model utilizes a supervised classification system to sequentially extract soil-landscape associations from digital data layers. Digital data serve as proxies for soil forming factors important in understanding the pedologic history and classification of soil in remote areas. Evidence is gathered through extensive field observations in order to understand the dominant pedologic processes and to establish the digital data threshold values. Mapping with RASP allows a soil scientist to execute and capture repeatable calculations and logic. Historically, this tacit knowledge was commonly available as a mental model developed by individual soil scientists. In many cases, the mental model was never adequately captured. To date, RASP has successfully mapped over 284,000 hectares of remote land in Washington State. Initially, RASP was used to map 254,000 hectares in the Pasayten and Sawtooth Wilderness. Mapping results have been correlated and incorporated into the soil survey of the Okanogan National Forest. Thunder Creek Watershed, 30,000 hectares within North Cascades National Park, has also been mapped with RASP. Ongoing development and validation of RASP is being undertaken with additional soil survey mapping in the three major National Parks in Washington State; North Cascades, Rainier, and Olympic. In order to better evaluate the effectiveness of RASP, statistical methods for determining reliability of delineations and agreement with field observations are underway. Initial analysis by direct comparison of field observations and RASP output revealed an 83% agreement, quite sufficient for fourth order surveys of remote areas. In many cases, smoothing of noisy input data and better GPS accuracy of observation sites improves the overall agreement. In addition to this direct comparison, we have also run unsupervised and supervised clustering algorithms to elucidate natural soil-landscape associations for comparison with RASP output. A regression analysis is also being run to bolster statistical validation and accuracy of data layer influence and sequencing in model construction.