Multiyear yield data may lead to more accurate determination of yield-soil relationships. This study was conducted to determine: if categorical yield classes could be found; if soil and/or topographical properties differed among yield classifications; and if soil and/or topographical properties could be used to classify field locations into yield classes. Soybean (Glycine max L.) yields over 4 yr in three fields were classified into four zones: consistent-high, consistent-average, consistent-low, and inconsistent. Soil samples from each zone were analyzed for pH, extractable nutrients, total C and N, texture, and elevation. Slope, plan curvature, and profile curvature at each sampling point were determined from elevation. While differences existed, there were no consistently different soil or topographical properties among yield classes in each of the three fields. Linear discriminant analysis found unique yield-affecting factors in each field. In the North field, sand, K, and pH were found to influence yield classification. In the South field, clay, sand, slope, Mg, and plan curvature predicted yield classification. In the East field, pH and clay significantly influenced yield classification. The derived functions classified sampling locations into appropriate yield classes 60 to 100% of the time. The most accurate prediction rate came when classifying the soil sampling points into the high yield zone in the East field. The highest error rate also occurred in this field when attempting to predict which points fell into the average classification. The combination of yield classification and linear discriminant analysis appears promising when trying to determine soil-topography-yield relationships and, hence soil management zones.