Locating the exact boundaries of soil map units is one of the primary objectives for soil surveyors. Statistical methods were used to assure the most accurate location. Soil spatial variability, autocorrelation function, and soil boundary locations were examined along a 2700-m transect in southern New Mexico. Eighty-nine observation points were equally spaced along the transect. Selected physical and chemical characteristics through the transect were determined. A mulitvariate method of principal-component analysis was used to produce one set of data. These data were first inspected for stationary manner, i.e., that the mean and variance of each property remain fairly constant for each data set. Log-normal transformation was used to detrend the data. The stationary manner of autocorrelations was tested with semivariograms. The range of dependence obtained from the autocorrelations and semivariograms was used in a squared-Euclidean-distance procedure to locate the soil boundaries. These boundaries were compared with those obtained by conventional soil-survey methods. Some of the calculated boundaries agreed with those obtained by conventional soil survey. The latter method is more economical and more productive than the statistical method.