In soil survey, mappers must recognize, record, interpret, and convey meaningful soil-geographic patterns. They accomplish this by amassing soil point data and classifying each pedon into its appropriate preestablished taxonomic unit, a process which largely is empirical, analytical, and deductive. They also create map units intended to depict soil patterns across the landscape. They do this by inferring and delineating perceived spatial changes in one or more of the soil-forming factors, a process which principally is rationalistic, predictive, and inductive. Concurrently, they contrive soil-landscape models to correlate and extrapolate point data into the spatial units. Mappers regard the process as a mix of science and art. Although they have been formally educated and trained in the sciences of soil morphology and classification, they rarely have been adequately schooled in the “art” of creating reliable spatial patterns from an assortment of pedons and taxonomic units. They proceed by trial and error, perhaps with capable mentoring, but with little written guidance. Three factors confound the situation and contribute to confusion. First, each of the five soil-forming factors has multiple attributes, resulting in dozens of choices with no firm criteria for setting and applying priorities. Second, the soil-landscape models remain mostly in the mappers' heads—at least until the survey is published. Although recent advancements address this shortcoming, a significant gap remains between concept and practice. Third, taxonomic names (e.g., soil series), which are determined deductively from elaborate, exacting, uniform criteria arranged in a strict written hierarchy (Soil Taxonomy), are applied to map units, which are created inductively from entirely different, unwritten and incoherent models. This paper proposes that the fundamental process of designing map units be transformed from its current rationalistic-predictive-inductive approach to one that parallels the empirical-analytical-deductive methods employed in taxonomic decisions. Map units should derive from their own clearly defined spatial criteria, rather than by assumptions of soil-forming factors and extrapolations from point data. This paper further proposes that rationalistic-predictive-inductive models be employed not to design map units, but solely (insofar as is practicable) to correlate taxonomic units (along with pedologic and edaphic data) with map units. Two related frameworks are put forward to advance the proposals. The first is a simple landscape hierarchy intended to parallel the taxonomic hierarchy. It is a rule-restricted four (or in some cases five) tiered ranking of spatially related natural and constructed features. It provides a qualitative foundation for map unit design and for sophisticated quantitative and technological applications. Second, a dichotomous key is offered to qualitatively structure soil-landscape models and link taxonomic units to map units. Finally, an alternative naming convention is suggested to reflect the criteria by which map units are designed. This system has been applied successfully in the soil survey of Yosemite National Park, California. Perhaps more importantly, students have shown no confusion in understanding or accepting the approach.