In order to evaluate the effects of reducing the number of animals used in the NIH mouse protection test for potency determination of inactivated rabies vaccines for human use, a retrospective study of the results obtained in the Brazilian National Control Laboratory, Instituto Nacional de Controle de Qualidade em Saúde (INCQS), was performed, comprising 214 vaccine lots. The INCQS Standard Operating Procedure establishes the use of three vaccine dilutions and 18 animals per dilution, separated into two cages with 9 mice each. The results of the two cages of each dilution were considered as two different groups (C1 and C2), and therefore, for each vaccine lot, three results were obtained: one for the standard test (ST) with 18 mice, one using the C1 cages with 9 mice and another using the C2 cages with 9 mice. The results were evaluated as repeated measures of the same method on the same samples. In this study, the effects of the reduction in: (a) the measurement error and its association with the size of measurement, (b) the agreement between the results using the concordance coefficient of correlation, (c) the agreement of categorized results as "Pass" or "Fail" using the Kappa index, (d) the precision of potency determinations using the 95% confidence interval and (e) the incidence of statistically invalid assays due to non-linearity and non-parallelism were evaluated. It was concluded that the results from the NIH mouse protection test using 9 mice per dilution are in good agreement with the results obtained using 18 mice per dilution. Therefore, nine animals per dilution is a suitable number to meet the statistical requirement for valid assays.