© The Author(s) 2018. Aims Relatively little is known about the health outcomes associated with very low plasma concentrations of high-density lipoprotein cholesterol (HDL-C) mainly because of the small numbers of individuals with such extreme values included in clinical trials. We, therefore, investigated the association between low and very low HDL-C concentration at baseline and incident all-cause-mortality, death from malignant disease (i.e. cancer), and with fatal or non-fatal incident coronary heart disease (CHD) in individuals from the Reasons for Geographical And Racial Differences in Stroke (REGARDS) study. Methods and results Analysis was based on 21 751 participants from the REGARDS study who were free of CHD, other cardiovascular disease, and cancer at baseline and were categorized by baseline HDL-C into <30 mg/dL (very low), 30-<40 mg/dL (low), and ≥40 mg/dL (reference). A series of incremental Cox proportional hazards models were employed to assess the association between the HDL-C categories and outcomes. Statistical analysis was performed using both complete case methods and multiple imputations with chained equations. After adjustment for age, race, and sex, the hazard ratios (HRs) comparing the lowest and highest HDL-C categories were 1.48 [95% confidence interval (CI) 1.28-1.73] for all-cause mortality, 1.35 (95% CI 1.03-1.77) for cancer-specific mortality and 1.39 (95% CI 0.99-1.96) for incident CHD. These associations became non-significant in models adjusting for demographics, cardiovascular risk factors, and treatment for dyslipidaemia. We found evidence for an HDL paradox, whereby low HDL (30-<40 mg/dL) was associated with reduced risk of incident CHD in black participants in a fully adjusted complete case model (HR 0.63; 95% CI 0.46-0.88) and after multiple imputation analyses (HR 0.76; 95% CI 0.58-0.98). HDL-C (<30 mg/dL) was significantly associated with poorer outcomes in women for all outcomes, especially with respect to cancer mortality (HR 2.31; 95% CI 1.28-4.16) in a fully adjusted complete case model, replicated using multiple imputation (HR 1.81; 95% CI 1.03-3.20). Conclusion Low HDL-C was associated with reduced risk of incident CHD in black participants suggesting a potential HDL paradox for incident CHD. Very low HDL-C in women was significantly associated with cancer mortality in a fully adjusted complete case model.