BACKGROUND: Sleep duration has been associated with depression. However, mean regression, such as linear regression or logistic regression, may not capture relationships that occur mainly in the tails of outcome distribution. This study aimed to evaluate the associations between sleep duration and depression along the entire distribution of depression using quantile regression approach. METHODS: This study included 55,954 adults aged 18 to 80 years from the National Health and Nutrition Examination Survey (N = 34,156) and the Korea National Health and Nutrition Examination Survey (N = 21,798). The coefficients corresponding to cross-group differences in PHQ-9 scores were estimated when comparing short or long sleep duration with normal sleep duration on deciles of PHQ-9 score distribution. RESULTS: At lower quantiles, either short or long sleep duration was not associated with depression. At higher quantiles, the association of both short and long sleep duration with depression became much more pronounced. Compared with normal sleep duration, short and long sleep duration were associated with increases of 1.34 (95 % CI: 1.16, 1.51) and 0.28 (95 % CI: 0.04, 0.52) in PHQ-9 scores at the 50th quantile, while the corresponding increases were 3.27 (95 % CI: 2.83, 3.72) and 1.65 (95 % CI: 0.86, 2.45) at the 90th quantile, respectively. We also found that the magnitude of association between short sleep duration and depression was stronger among females and individuals with chronic diseases. CONCLUSIONS: The beneficial effect of sufficient sleep in decreasing depression severity may be more evident among individuals with severe depression. Further studies could explore whether these heterogeneous associations can be generalized to populations with different characteristics.