Atemoya currently has its seeds studied in several aspects, from the technological and physiological point of view. However, for the performance of the germination test, there is no standardization in relation to the number of seeds and replicates. Thus, this work aims to determine the optimal sample size for germination tests with atemoya seeds. A germination test was carried out with 5 treatments, considering 10, 20, 30 40 and 50 seeds per sampling unit with 50 replicates of each. Data were analyzed using logistic regression, non-linear Gompertz regression models, bootstrap simulation and graphs in the form of contour lines in order to be able to infer the best binomial of the number of seeds per sampling unit and the number of replicates. The expected germination percentage, maximum germination speed and times for the beginning of the germination process, maximum germination speed and interruption of the germination process were determined. The treatment of 20 seeds per sampling unit statistically differs from that of 40 seeds by the Tukey-Kramer test with 5% significance. It could be concluded that for germination tests with atemoya seeds to have statistical validity, the smallest sample size is 10 seeds per unit, regardless of number of replicates.