In this study, we applied multilevel piecewise linear regression with a random-effects model under a Bayesian estimation framework to a total of 5,072 7th grade students in 35 schools to search for an optimum time in terms of the effect of homework time on student academic achievement. First, we found a nonlinear relationship between time spent on homework and student academic achievement. Second, we found an optimum duration of homework with two rather distinguishing phenomena of patterns while controlling only student socioeconomic status (SES) and gender. The optimum time spent on homework is 1.77 hours. However, after further controlling students' prior academic achievement, the optimum time disappeared. We discuss possible explanations for the disappearance of the optimum time and the implications for further research directions.