An experimental design in which the number of treatments is the same as the number of replications, and each treatment occurs once in every column and row. It is analagous to a *randomized block
design in two directions. It is often used in fertilizer field trials and has the advantage of eliminating from the total variation environmental differences, such as soil fertility, that exist across and down the square experimental plot. The Latin square generally yields more useful information that would a randomized block design of similar size. However it has the limitation that with a large number of treatments there must consequently be a large number of replicates and beyond a certain point the labour involved is not worth the information obtained. Also, with small squares, the method is insensitive because the number of degrees of freedom is low. This can be overcome by replicating the squares.