Terrain analysis of Digital Elevation Model (DEM) data has become important for assessing soil variability. Such soil-landscape modeling techniques are widely used as a quantitative method to predict patterns of soil properties from DEM and other environmental variable data. However, the scale of analysis influences calculated terrain attributes and resulting soil-landscape relationships. This occurs because the processes that control soil variability, and consequently the landscape factors that regulate these processes, vary at different spatial scales. Selecting a single DEM resolution to model soil variability is therefore highly dependent on the scale of the process being modeled. These choices are accentuated by the development of new DEM data sources, particularly those at exceptionally high spatial resolutions (e.g., LIDAR). Attempts to determine an ideal DEM resolution have found varied optima, from <5 m to >30 m, depending on various soil and landscape factors. Resampling and generalizing a high-resolution DEM to achieve a lower resolution product for calculating terrain attributes at coarser scales may better approximate important soil-landscape relationships but also results in an unnecessary loss of information. An alternative may be to calculate terrain attributes from a high-resolution DEM, but at longer lag distances. We developed an object-oriented Microsoft Windows-based multi-scale terrain analysis application based on second-order finite difference solution applied to a 3x3 grid of elevation values. Unlike standard algorithms used for terrain attribute calculation, this application allows for this 3x3 moving window to use non-adjacent grid cells that are separated by a user-determined lag distance. The application was developed with a modular design for 2-D visualization and export of gridded terrain attributes, including slope gradient, slope aspect, and various measures of slope curvature (total, profile, contour, and tangential). We used this application with multiple sources of high-resolution DEM to (i) examine scale effects on terrain attribute values, and (ii) examine scale effects on soil-landscape relationships. The increasing availability of LIDAR data is providing opportunities for wide-area studies using high-resolution DEM data. We sought to quantify changes in terrain attribute characteristics when calculated at different scales from the same high-resolution data set. Results indicate that as lag distance increases, calculated slope gradients decrease (become less steep) and curvatures approach zero (become flatter), with statistically significant differences in terrain attribute distributions in all cases. These results mirror those found when comparing separate DEM of differing resolutions. Spatially, as lag distances increase, while there is a general smoothing of the topographic surface that leads to the loss of microtopographic features there is also a decrease in short-range variability such that areas of similar terrain attribute values aggregate to expose distinct landscape features, such as ridges and valleys. Precision surveying methods, such as real-time kinematic GPS, have lead to rapid production of high-resolution DEM data. We sought to examine correlations between terrain attributes and selected soil properties associated with changes in the scale of terrain attribute calculation. Correlation coefficients between terrain attributes and selected soil properties are sensitive to the lag distance. More specifically, correlations between selected soil properties and terrain attributes are not always highest when calculated over the shortest lag distance from a high resolution DEM. Slope gradient is most correlated with selected soil properties at shorter lag distances (5-10 m). Slope curvature is always most correlated with selected soil properties at lag distances greater than 10 m. Correlation values between selected soil properties and slope curvature are more sensitive to lag distance. When developing multivariate statistical soil-landscape models, our results indicate that including terrain attributes calculated at longer lag distances improves the quality of regression models. However, no single lag distance was best for predicting soil property variability.