Numerical classification of soils was studied with emphasis on methodology and feasibility in survey. A procedure was designed for construction of classes sufficiently homogeneous in terms of relevant properties and handlable by the surveyor. In the procedure 'central' depth-profiles are calculated separately for each property (e.g. clay content), from a sample of depth-profiles, with a relocation method minimizing within-class variances. Any soil profile can thus be identified in the field by allocating its constituent depth-profiles to the central depth-profile that is most similar for the respective properties. Resulting strings of class labels serve for interim data recording. If too many combinations of central depth-profiles arise to map all individually, they are fused into larger classes and within-class variances are again minimized. This procedure was applied to survey data from a marine clay area in the Netherlands: field estimates for 6 properties in 2212 profiles divided into 20 depth intervals. A new method was used to map classes automatically. Tests showed that: samples of several hundred profiles were needed; order of profiles and initial solution for relocation had little effect on results; only extreme weighting significantly affected homogeneity for different variables. Choice of weights and number of classes should be related and supported by sensitivity analysis..

Part I: Cultural practices; Part II: Quality of the grain and improvement of the crop by breeding.