Soil sampling using a 67 m x 67 m grid was carried out at four locations in conjunction with a precision (site-specific) farming study. The four locations were on undulating to rolling topography, distributed across the Brown, Dark Brown and Black soil zones of the plains region of western Canada. A Differential Global Positioning System (DGPS) was used for positioning to map yield, terrain, fertility, and salinity. These attribute maps were used to develop maps for variable rate fertilization. At each grid-line intersection (node), a composite sample consisting of 15 cores was taken within a 5 m radius. Sample depths were 0-15, 15-30, 30-60, and 60-90 cm. The nutrient levels within the areas sampled (15-25 ha) generally had a wide range, high standard deviation, and strong positive skewness. The large variations found in nutrient levels and crop yields support the need for variable rate fertilization. For conventional, constant rate fertilizer application, the strong positive skewness of nutrient levels obtained with grid sampling indicates that systematic errors occur with the current method of composite sampling of large fields. When grid sampling results have frequency distributions that are positively skewed, fertilizer recommendations based on composite sampling will under fertilize the majority of the field.