In contrast to conventional conditional mean approaches, this paper implements the quantile regression, both parametrically and semiparametrically, to re-examine the validity of the Kuznets hypothesis across different quantiles of the conditional inequality function. Our empirical results show that an inverted-U-shaped relationship between inequality and per capita GDP prevails in countries with mild income inequality, but not for ones with too high or too low income inequality. These results are robust to the use of different data sets, controlling variables and model specifications.