The purpose of this study is to find statistically-significant number of subjects in a driving experiment that investigates the level of passenger acceptance of pavement roughness. Statistical power analysis is used to determine an appropriate sample size as a function of a significant level, effect size (ES) and statistical power. In power analysis, the ES basically depends on the purpose of each individual study. Therefore, we estimate the ES to evaluate the acceptance ratio of roughness based on a subjective rating by a driving simulator. Consequently, over 20 subjects in the experiment allows one to compare the acceptable proportion of two vertical accelerations split by more than 0.35m/s2 in the root-mean-square value that is medium or large ES. Conversely, the small ES offered by the acceleration difference of less than 0.3m/s2 requires a large number of subjects such as 400. If the required number of subjects is not obtained, we recommend that the statistical confidence interval should be calculated and the result of the experiment should not be generalized for the road surface evaluation.