Employing a rebaselining concept may reduce noise in retinal layer thinning measured by optical coherence tomography (OCT). From an ongoing prospective observational study, we included patients with relapsing multiple sclerosis (RMS), who had OCT scans at disease-modifying treatment (DMT) start (baseline), 6-12 months after baseline (rebaseline), and ⩾12 months after rebaseline. Mean annualized percent loss (aL) rates (%/year) were calculated both from baseline and rebaseline for peripapillary-retinal-nerve-fiber-layer (aLpRNFLbaseline/aLpRNFLrebaseline) and macular-ganglion-cell-plus-inner-plexiform-layer (aLGCIPLbaseline/aLGCIPLrebaseline) by mixed-effects linear regression models. We included 173 RMS patients (mean age 31.7 years (SD 8.8), 72.8% female, median disease duration 15 months (12-94) median baseline-to-last-follow-up-interval 37 months (18-71); 56.6% moderately effective DMT (M-DMT), 43.4% highly effective DMT (HE-DMT)). Both mean aLpRNFLbaseline and aLGCIPLbaseline significantly increased in association with relapse (0.51% and 0.26% per relapse, p < 0.001, respectively) and disability worsening (1.10% and 0.48%, p < 0.001, respectively) before baseline, but not with DMT class. Contrarily, neither aLpRNFLrebaseline nor aLGCIPLrebaseline was dependent on relapse or disability worsening before baseline, while HE-DMT significantly lowered aLpRNFLrebaseline (by 0.31%, p < 0.001) and aLGCIPLrebaseline (0.25%, p < 0.001) compared with M-DMT. Applying a rebaselining concept significantly improves differentiation of DMT effects on retinal layer thinning by avoiding carry-over confounding from previous disease activity.