Event-related potentials (ERPs) offer unparalleled temporal resolution in tracing distinct electrophysiological processes related to normal and pathological cognitive aging. The stability of ERPs in older individuals with a vast range of cognitive ability has not been established. In this test-retest reliability study, 39 older individuals (age 74.10 (5.4) years; 23 (59%) women; 15 non β-amyloid elevated, 16 β-amyloid elevated, 8 cognitively impaired) with scores on the Montreal Cognitive Assessment (MOCA) ranging between 3 and 30 completed a working memory (n-back) test with three levels of difficulty at baseline and 2-week follow-up. The main aim was to evaluate stability of the ERP on grand averaged task effects for both visits in the total sample (n = 39). Secondary aims were to evaluate the effect of age, group (non β-amyloid elevated; β-amyloid elevated, cognitively impaired), cognitive status (MOCA), and task difficulty on ERP reliability. P3 peak amplitude and latency were measured in predetermined channels. P3 peak amplitude at Fz, our main outcome variable, showed excellent reliability in 0back (intraclass correlation coefficient (ICC), 95% confidence interval = 0.82 (0.67-0.90) and 1-back (ICC = 0.87 (0.76-0.93), however, only fair reliability in 2-back (ICC = 0.53 (0.09-0.75). Reliability of P3 peak latencies was substantially lower, with ICCs ranging between 0.17 for 2-back and 0.54 for 0-back. Generalized linear mixed models showed no confounding effect of age, group, or task difficulty on stability of P3 amplitude and latency of Fz. By contrast, MOCA scores tended to negatively correlate with P3 amplitude of Fz (p = 0.07). We conclude that P3 peak amplitude, and to lesser extent P3 peak latency, provide a stable measure of electrophysiological processes in older individuals.