Background: Chordomas are rare malignant skull-base/spine cancers with devastating neurological morbidities and mortality. Unfortunately, no reliable prognostic factors exist to guide treatment decisions. This work identifies DNA methylation-based prognostic chordoma subtypes that are detectable non-invasively in plasma. Methods: Sixty-eight tissue samples underwent DNA methylation profiling and plasma methylomes were obtained for available paired samples. Immunohistochemical staining and publicly available methylation and gene expression data were utilized for validation. Results: Unsupervised clustering identified two prognostic tissue clusters (log-rank p=0.0062) predicting disease-specific survival independent of clinical factors (Multivariable Cox: HR=16.5, 95%CI: 2.8-96, p=0.0018). The poorer-performing cluster showed immune-related pathway promoter hypermethylation and higher immune cell abundance within tumours, which was validated with external RNA-seq data and immunohistochemical staining. The better-performing cluster showed higher tumour cellularity. Similar clusters were seen in external DNA methylation data. Plasma methylome-based models distinguished chordomas from differential diagnoses in independent testing sets (AUROC=0.84, 95%CI: 0.52-1.00). Plasma methylomes were highly correlated with tissue-based signals for both clusters (r=0.69 & 0.67) and leave-one-out models identified the correct cluster in all plasma cases. Conclusions: Prognostic molecular chordoma subgroups are for the first time identified, characterized, and validated. Plasma methylomes can detect and subtype chordomas which may transform chordoma treatment with personalized approaches tailored to prognosis.