Objective We aimed to re-examine mortality risk estimates for metabolically healthy obesity by using a ‘stable’ healthy non-obese referent group. Design Prospective cohort study. Methods Participants were 5427 men and women (aged 65.9 ± 9.4 years, 45.9% men) from the English Longitudinal Study of Ageing. Obesity was defined as body mass index ≥30 kg/m² (vs non-obese as below this threshold). Based on blood pressure, HDL cholesterol, triglycerides, glycated hemoglobin and C-reactive protein, participants were classified as ‘healthy’ (0 or 1 metabolic abnormality) or ‘unhealthy’ (≥2 metabolic abnormalities). Results Totally, 671 deaths were observed over an average follow-up of 8 years. When defining the referent group based on 1 clinical assessment, the unhealthy non-obese (hazard ratio (HR) = 1.22; 95% CI: 1.01, 1.45) and unhealthy obese (HR = 1.29; CI: 1.05, 1.60) were at greater risk of all-cause mortality compared to the healthy non-obese, yet no excess risk was seen in the healthy obese (HR = 1.14; CI: 0.83, 1.52). When we re-defined the referent group based on 2 clinical assessments, effect estimates were accentuated and healthy obesity was at increased risk of mortality (HR = 2.67; CI: 1.64, 4.34). Conclusion An unstable healthy referent group may make ‘healthy obesity’ appear less harmful by obscuring the benefits of remaining never obese without metabolic dysfunction.