This study aimed at determining the optimal number of replications and measurements necessary for the screening of soybean cultivars subjected to three conditions of water availability during the initial stage of development. Two experiments were carried out in a greenhouse. A randomized block design was used, in a factorial scheme with two factors (cultivar and water availability), six replications, ten cultivars and three conditions of water availability (control condition, in which the sand was moistened with water; stress control in the water potential of -0.2 MPa and stress condition in the water potential of -0.4 MPa). At 20 days after cultivation, measurements were taken, subsequently estimating the repeatability coeffi cient, of the number of necessary measurements and of the determination coeffi cient. The repeatability coeffi cient was estimated by distinct statistical procedures, such as method of analysis of variance, principal components with basis on the matrices of phenotypic variances and covariances; principal components with basis on the matrices of correlation and structural analysis. For the determination of the number of replications, the repeatability coeffi cients ranged from low to high in experiment I, except for the root variables, which presented high values in both experiments. In experiment II, in turn, the magnitudes of the determination coeffi cients were above 90%. For plant height, number of leaves, root length and dry matter of the aerial part, only two replications would be necessary to obtain 80% confi dence. Regarding the number of measurements per parcel, only one plant could be evaluated in each parcel, with 85% confi dence for both experiments. Only two replications and one measurement per parcel are enough for high estimates of repeatability. This way, adequate effi ciency is obtained in the selection of superior materials for tolerance to water defi cit, during the initial stages of soybean development.