Various methods are considered for controlling the environmental variation over the area of an experiment. The conventional way is to use blocks or, alternatively, rows and columns. Other possibilities are the fitting of a paraboloidal surface and adjusting each plot by the performance of its neighbour which last has several variants. Data were taken from five series of experiments conducted in Africa and the West Indies on maize, groundnuts and cowpea. Each body of data was analysed using each of the methods. Blocks were disappointing in their effectiveness but rows and columns were very successful. The fitting of paraboloids gave encouraging results and so did adjustment by neighbouring plots, especially when an iterative solution was sought. With all methods there was a tendency for control to be more effective when experimental error was less. Unfortunately the most effective methods, i.e., the use of rows and columns and adjustment by neighbouring plots, are both open to objection, the former because it is restricted by the difficulties of row-and-column designs, which are understood but complicated, and the latter by lack of the basic mathematical theory. Also, in the present study some disconcerting features were noted, which need to be examined further.