Aim Studies have shown better predictive value of self‐rated health (SRH) for mortality when prospective change in SRH is considered. However, retrospective change is more feasible and might have better sensitivity to objective health changes. This study compares the predictive value for mortality of retrospectively measured change in SRH (based on a “then‐test”) with current SRH and prospectively measured change in SRH. Methods Data from two waves of the Longitudinal Aging Study Amsterdam (2001–2003 and 2005–2006 [T0], n = 1894) were used. Retrospective change was defined as the difference between SRH at T0 (“current SRH”) and SRH measured with a then‐test at T0, asking for a renewed judgement of one's health at the previous wave. Prospective change was defined as change in SRH between the two waves. We applied Cox proportional hazards analysis to predict 5‐year mortality. Results Having poorer current SRH significantly predicted mortality (HR poor vs very good SRH = 4.42). Declined SRH was associated with higher mortality risk, but only when measured prospectively (one point decline vs no change HR = 1.33; two points decline HR = 1.95). After adjusting for current SRH, neither change measure predicted mortality. Results were similar in subgroups that did and did not experience incident diseases or limitations between the two waves. Conclusions Neither retrospective, nor prospective changes in SRH improved the prediction of mortality in older adults over current SRH. These results imply that using a standard single indicator for self‐rated health in research or clinical practice might suffice to identify those with a high risk of future negative health outcomes. Geriatr Gerontol Int 2013; 13: 678–686.