Although all methods of planning for the maximization of farm profits require knowledge of the basic ‘input-input’ and ‘input-output’ relationships, particularly detailed information is required when using linear programming. Specially designed crop experiments are one major source of such information, but it is important that each experiment should include all major interacting variables, since the response curves estimated from the data, and the relationships extracted from these curves, will otherwise be of little use for determining economic optimum input combinations. Where the matter of interest is the most likely production response, under the planning conditions envisaged, the best available estimate will usually be the average response estimated from a sample of experiments distributed randomly within the conditions. In each experiment of such a sample, the variables being examined should be at the same series of input levels, whilst other variables which are not of interest, but can be controlled, should be held constant at the same general level. With respect to the design of experiments established to generate production relationships, complete factorial and composite or rotatable arrangements are considered to be most suitable, although each of the types has certain disadvantages. Regarding the derivation of ‘input-input’ and ‘input-output’ relationships from the response curve fitted to the data obtained from such experiments, it is felt that this curve should first be ‘scaled down’, to allow both for the more favourable experimental conditions and for losses between the field and the point of use. After this, a number of individual production relationships should be selected, on that portion of the curve expected to be of economic interest. The relationships can then be employed in a linear programming procedure designed to generate a farm plan incorporating that combination of enterprises likely to earn the maximum profit.