Abstract. A number of different types of edge effects on experimental plots are defined, and
it is shown that they are frequently important in agroforestry experiments. Methods of
eliminating edge effects, either at the design or the analysis stage, are described. Computer
simulations, based on uniformity data, are used to investigate the efficacy of neighbourbalanced designs, with and without neighbour terms, for reducing the problem of neighbour
effects between different treatments. It is concluded that, while neighbour-balanced designs
may be useful in certain situations or in combination with other methods, it is preferable to
eliminate edge effects by using adequate guard rows. Recommendations are therefore made
concerning the correct use of guard areas in agroforestry experimentation.
Introduction