Abstract This study intends to estimate the coefficient of repeatability for traits relating to yield, so as to infer what is the minimum number of evaluations required to identify superior genotypes; to verify adaptability and stability; to predict genetic gains; and to select the best families and in S1 progenies of Psidium guajava. Families were obtained by means of self-fertilization of superior genotypes (full sibs) from the guava tree genetic breeding program of the UENF. The experiment was implemented at the Experimental Station of Ilha Barra do Pomba, in the municipality of Itaocara-RJ. In this case, an experimental randomized block design was used with 18 inbred families, three replicates, and ten plants per plot. Inbred families were evaluated in three harvests regarding fruit weight, number of fruits per plant, and total yield. It was then conducted an analysis of repeatability, adaptability, and stability of genotypes by means of the Selegen/REML/BLUP program. Repeatability values show that the performance of genotypes is constant among measurements, suggesting the need to perform only three evaluations for the selection of superior genotypes. There was agreement on the choice of families 1, 4, 6, and 12 as the most yielded, adaptable, and stable for the traits evaluated.