Introduction: We present the dosimetric evaluation of craniospinal irradiation (CSI) treatments delivered with protons at Paul Scherrer Institute (PSI), with special focus on local recurrences and late toxicity outcome. Methods: This study included 71 children, adolescents and young adults (c-AYA), who received or intended to receive (3 patients, pts) CSI using PBS-PT at PSI between 2004 and January 2021. The most frequent primary tumours were: medulloblastoma (42 pts), ependymoma (8 pts) and germ cell tumors (6 pts). The patients were treated prone on Gantry1 (G1; 22 pts) up to 2017, and afterwards supine on Gantry2 (G2; 49 pts). Accuracy of prone vs. supine setup was evaluated. Nine patients received CSI for local failure (LF) after a first course of local fractionated radiation therapy (RT). For 59/71 patients (excluding three patients not receiving PBS-PT CSI and nine preirradiated) CSI plans were compared considering gantry and planning technique. Detailed analysis of the full treatment (CSI and boost series) was performed for 8 patients presenting with LFs (4 of them presented also distal failure) and for selected patients presenting with late toxicity (G2 to G4) or asymptomatic radiation-induced radiological findings. Results: Supine positioning resulted in lower systematic and random errors as compared to prone (0.25 mm and 0.4 mm systematic errors respectively for supine and prone; random errors in PA direction reduced from 1.8 mm for prone to 1.4 mm for supine). Conclusions: LFs were not correlated with potential dose inaccuracies or lack of robustness and no correlation of toxicities to enhanced LET have been observed.