In any experiment the land will not be completely uniform and it is helpful to have ways of allowing for the variation. Some sources, for example altitude, will be obvious; some will be permanent but less obvious, such as depth of topsoil; and some may depend upon season and so be unpredictable. Various methods of local control have been suggested. Where the experimenter has good knowledge of the land it can be divided into blocks, each as far as possible uniform within itself. Then comparisons are made within the blocks rather than within the area as a whole. Where such knowledge does not exist, it is sometimes reasonable to make an assumption about the fertility pattern and make use of that, as in a row-and-column design like a Latin square. There is also the possibility of judging the fertility pattern from the data themselves and assessing the performance of a plot by reference to that of its neighbours.

The approach will be to generate bodies of data on the computer to form realizations of diverse fertility patterns and to use all methods on all realizations, noting success and failure.

When the variation forms a trend, blocks succeed only if they are aligned along fertility contours; the other methods do not depend upon orientation. Row-and-column designs can fail badly if the rows and columns interact. Some random variation is inevitable and it makes all methods less effective, especially nearest-neighbour methods, which can fail also when there are discontinuities. Random patches of different soil types are very difficult to deal with and any method might fail.