The conditions required for a valid analysis of variance are enumerated. A 2x4 factorial experiment with corn in randomized blocks, designed to study data of sowing and types of planting for corn, is discussed. The experiment was performed at the Estação Experimental de Serrinha, Estado da Bahia. The original data showed that the variances within treatments were not homogenous. It was found by means of a graphical presentation that the relation - standard deviation / mean of treatment - was almost constant for the different treatments. A log transformation brought the desirable homogeneity to the variances ; the usual analysis of variance was then carried out.