Samples of both forest floor and surface mineral soil (0–25 cm) were collected from plots in 171 forest stands across Minnesota, Wisconsin, and Michigan. Five forest types were represented: Abies balsamea (L.) Mill., Pinus banksiana Lamb., Pinus resinosa Ait., Populus tremuloides Michx., and northern hardwoods dominated by Acer saccharum Marsh. Samples were analyzed for pH, total N and S, loss on ignition, and exchangeable Ca, Mg, K, Na, and acidity; the latter were summed to yield cation exchange capacity. Data for each variable included 40 pairs of laboratory duplicates; 45 samples representing three sampling points approximately 2.5 m apart within a subplot in each of 15 plots (spatial scale of 100.4m between samples); and 830 samples from five subplots approximately 30 m apart within the 171 plots (scale of 101.5m). Variation among samples was determined for each of those categories, as well as among the nearly 35 plots within a forest type at approximately 100 km apart (scale of 105m), and among the five forest types. Increase in variation with spatial scale was statistically significant. Coefficients of variation were very similar to those from other studies; considering the range of locations, soils, and vegetation in the literature, reported coefficients of variation for soil properties are remarkably uniform. None of the variables was normally distributed, even after log or algebraic transformations. Determination of number of samples necessary to characterize a system adequately is fraught with uncertainty regarding assumptions of the form of the distributions and of the level of precision desired. The commonly accepted precision of 10% of the mean from a normal distribution at a defined confidence level needs re-evaluation.