Optimized agricultural management depends on reliable and detailed information describing the spatial distribution of soils, geology (parent material), and topography. Subsurface attributes (earth data) have in common that they vary through three-dimensional (3-D) space and through time. Our goal was to overcome current limitations of geographic information systems to manage, analyze, and visualize geographic data in two dimensions (2-D). In this paper we present an approach to reconstruct and visualize soil-landscapes with focus on web-based dissemination of model output. Our approach is multi-dimensional, multi-variate, spatially object-oriented, transferable, scalable, and expandable. We used geo-referenced subsurface and topographic attributes from several sites in southern Wisconsin and northwestern Ohio to demonstrate the capabilities of our approach. Reconstruction was based on 2-D and 3-D ordinary kriging utilizing Environmental Visualization System (EVS) software and implemented in Virtual Reality Modeling Language (VRML). Three different types of soil-landscape models were distinguished: (i) Models representing subsurface attributes as points (ii) Models representing subsurface objects as polyhedrons or "volume models" and (iii) Block models consisting of voxels (volume cells). A server hosts the virtual soil-landscape models, which are accessible by multi-clients via an interface coded in HTML. These models are interactive, platform independent and enable users to analyze, explore and gain insight into the spatial distribution of topographic and subsurface attributes in 3-D geographic space. Numerous agricultural tasks are supported using our approach including 3-D soil surveys, informed decision-making, assessment of environmental quality, farm management, land use planning, and many more. Virtual soil-landscape models are beneficial in disseminating geo-referenced earth data to educators, researchers, government agencies, and the general public.