Introduction RTOG 1205 is the only randomized study to evaluate the safety and efficacy of reirradiation (reRT) in recurrent glioblastoma (GBM). While this study showed that reRT was safe and improves progression-free survival (PFS), an improved approach to reRT is still needed. In this study, we report on patterns of failure and outcomes in a cohort of patients with recurrent GBM who underwent reRT. We hypothesize that patients at high risk of leptomeningeal spread (LMS) are not good candidates for reRT due to the risk of treatment-related toxicity without clinical benefit. Methods In this retrospective study, patients with recurrent GBM who underwent reRT at a single institution from 2015-2023 were included. Sociodemographic, treatment, and outcomes data were collected via chart review. Time to progression was defined as the time from the start of reRT to progression per the Response Assessment in Neuro-Oncology (RANO) criteria. Overall survival (OS) was defined as the time from the start of reRT to death. PFS and OS were estimated using the Kaplan-Meier method. Results Thirteen patients with recurrent GBM who underwent reRT were identified. The median age at diagnosis was 58 years. Six patients (46.2%) had tumors that were O6-methylguanine-DNA methyltransferase (MGMT) methylated, four (30.8%) were MGMT unmethylated, and three (23.11%) had unknown MGMT status. Eight patients underwent repeat resection after recurrence and before reRT. Most patients (n=7) received 35 Gy in 10 fractions with concurrent bevacizumab, while other patients were treated with 25-40 Gy in 5-15 fractions with grade 1 or less acute toxicity. Three patients were treated with tumor-treating fields. The median follow-up was five months. Median PFS was three months [95% confidence interval (95% CI): one to four months] and median OS was five months (95% CI, 1-8 months) as compared to 7.1 months and 10.1 months, respectively, on RTOG 1205. Five patients developed LMS after reRT, one patient died before progression, and the remaining seven patients all developed progression within one centimeter of the recurrent tumor. Of the patients who developed LMS, all had tumors abutting the ventricles and three underwent resection 2-17 months before reRT. Conclusion Patterns of failure suggest a potential treatment selection approach for patients with recurrent GBM, in which patients at high risk of LMS (tumor abutting ventricles with or without recent surgery) should not undergo reRT, while patients at low risk of LMS are good candidates for reRT. Furthermore, reRT could be administered with reduced margins given that all non-LMS recurrences were within 1cm of the original tumor. Additional studies are needed to validate this approach.