Background: Magnetic resonance imaging (MRI) studies of multiple sclerosis–related fatigue had limited reproducibility. Temporal fatigue fluctuations have not been considered. Objective: To investigate whether a novel group allocation that reflects temporal dynamics of fatigue improves our ability to detect fatigue-associated structural brain abnormalities. Methods: Patient stratification based on biennial fatigue assessments: sustained fatigue (SF, n = 29, fatigued at the latest ⩾2 assessments), one time-point fatigue (1F, n = 15, fatigued at the latest, but non-fatigued at the penultimate assessment), reversible fatigue (RF, n = 31, non-fatigued at the latest assessment, but reported fatigue previously), and never fatigued (NF, n = 54). Brain parenchymal fraction (BPF) and T2 lesion volume (T2LV) were compared between these groups and were derived using a conventional, single time-point fatigued versus non-fatigued stratification. Results: The SF versus NF stratification yielded improved power. SF ( p = 0.005) and RF ( p = 0.043) showed significantly higher T2LV than NF. T2LV showed no significant differences in SF versus 1F, SF versus RF, or 1F versus RF. Fatigued versus non-fatigued patients showed significantly higher T2LV ( p = 0.030). We found no significant differences in BPF between the groups. Conclusion: Taking into account temporal fatigue dynamics increases the statistical power with respect to T2LV and may improve characterization of brain pathological correlates of MS-related fatigue.