Order 3 soil surveys are used by ranchers in Far West Texas to make stocking rate and brush management decisions. Despite the low intensity of information needs, the required soil data are expensive to acquire directly. Soil surveyors partially overcome high costs by utilizing proxy relationships, whereby relatively inexpensive data serve as surrogates for more costly information. Discovery of the specific relationships between particular soils and environment factors necessarily precedes their use to map soils in the field. The reconnaissance process for vector-based soil surveys is here termed “legend development”, and “discovery” for raster soil surveys. Regardless of the name, the process uses purposive sampling and inductive logic to develop general rules from a limited number of specific cases. The objective of legend development is to establish a legend, i.e., a list of map units. Map units are repetitive geographical soil bodies with reasonably uniform capability to support plant growth and engineered structures. Legend development uses gradsect (gradient directed transect) sampling of a small number of slope sequences, exploiting knowledge of geology, landforms, and vegetation. The process detects, evaluates, and ranks boundaries first, then subsequently describes and classifies soils between boundaries. For Order 3 soil surveys, map unit boundaries and identity are typically predicted from airphoto signature and surface features such as parent material, vegetation, and landform, then are periodically tested by direct examination. Vector soil survey is a divisive operation, partitioning an area into progressively smaller and more homogeneous units. In Culberson County, Texas, boundaries were identified that separate the Castile Formation from the Rustler Formation above and the Bell Canyon Formation below. The heterogeneous Castile terrain was then subdivided into conceptual segments each with less variable interpretations. Two multi-taxa map units (Pokorny-Dellahunt complex on summits and Joberanch-Dellahunt complex in drains) and two mono-taxa map units (Pokorny on hillsides and Elcor on breaks) were tentatively defined. The objectives of discovery differ somewhat from those of legend development. Whereas legend development establishes soil map units, discovery quantifies relationships between individual soil classes and the environment for use in predicting soil occurrence with SoLIM. In addition to understanding the landscape, discovery requires that the soil scientist explicitly and quantitatively express relationships between soil class and environmental factors. Potential data sources for soil inference include digital representations of relief, geology, and imagery. Discovery identifies tacit points where a soil class exists with a high degree of confidence. Membership within the target soil class for that grid cell is assumed to be unity. Georeferenced training data collected at the tacit point include slope gradient, plan curvature, and profile curvature, as well as reflectance (expressed by imagery digital numbers). Discovery deals with the landscape fabric, i.e. polygon boundaries and interiors, in reverse order as legend development; it focuses first on defining soil class in core areas, then secondarily on spatial transitions between classes. Relationships between soil classes and environmental factors were archived in a knowledge-base document. A unique combination of environmental parameters was established for each soil class. Four soil classes on the Castile Formation were related to slope gradient and profile curvature. Elcor breaks occur on slopes greater than 8 %, whereas gradients of Pokorny hillsides are between 2 and 8 %. Low (<2%) slope gradients characterize both Dellahunt benches and Dellahunt drains. Dellahunt benches are linear to slightly convex, whereas Dellahunt drains are concave. Alternatively, relative elevation or digital imagery brightness can be used to separate the Dellahunt soils on different landforms. Mathematically expressed soilscape models allow rapid creation of soil surveys by raster techniques. SoLIM is an inference engine that estimates soil class membership of each grid cell in the landscape based on the point-specific soil-environment relationships identified at tacit points. Through the discovery and implementation of quantitative soil-environment relationships, raster soil surveys have the potential to be more consistent and more accurate than polygon-based surveys created by traditional methods.