Abstract Background and purpose A constant relative biological effectiveness (RBE) is used for clinical proton therapy; however, experimental evidence indicates that RBE can vary. We analyzed pediatric ependymoma patients who received proton therapy to determine if areas of normal tissue damage indicated by post-treatment image changes were associated with increased biological dose effectiveness. Material and methods Fourteen of 34 children showed T2-FLAIR hyperintensity on post-treatment magnetic resonance (MR) images. We delineated regions of treatment-related change and calculated dose and linear energy transfer (LET) distributions with Monte Carlo. Voxel-level image change data were fit to a generalized linear model incorporating dose and LET. Cross-validation was used to determine model parameters and for receiver operating characteristic curve analysis. Tolerance dose (TD50 ; dose at which 50% of patients would experience toxicity) was interpolated from the model. Results Image changes showed dependence on increasing LET and dose. TD50 decreased with increasing LET, indicating an increase in biological dose effectiveness. The cross-validated area under the curve for the model was 0.91 (95% confidence interval 0.88–0.94). Conclusions Our correlation of changes on MR images after proton therapy with increased LET constitutes the first clinical evidence of variable proton biological effectiveness.