Objective <p>The EQ-5D is increasingly being used in studies of health inequalities, providing further evidence of a social gradient in health; i.e. consistent positive associations between a socioeconomic indicator and health. However, the steepness in the social gradients in HRQoL differs depending on which of the two EQ-5D measures is used; whether based on respondents’ EQ-5D-5L <i>descriptions</i> or by their <i>direct valuation</i> in EQ-VAS. This study aims to provide new knowledge as to why the two HRQoL measures suggest different degrees of health inequalities.</p> Methods <p>Based on two large unique data sets (Tromsø Study Wave 7, N = 21,083; MIC study, N = 8022), cross-sectional analyses were conducted. We identified the most prevalent EQ-5D-5L profiles. Within each of ten EQ-5D-5L profile groups, we examine response heterogeneities in EQ-VAS scores using linear regressions, as explained by respondents’ level of educational attainment, controlling for age and sex.</p> Results <p>We showed significantly increasing EQ-VAS scores along with educational attainments. For instance, in the most prevalent health state (11121), a consistent education-health gradient was observed: compared to individuals with primary education, the EQ-VAS was 1.7 higher among those with secondary education; 2.6 higher among individuals with short tertiary education; and, 3.7 higher among individuals with long tertiary education.</p> Conclusions <p>This paper provides new insights into the use of EQ-5D in health inequality studies by suggesting an additional underlying education gradient in HRQoL than what is revealed through the EQ-5D-5L values. Broader psychosocial domains and aspects of adaptation should be considered when monitoring health inequalities.</p>