A DNA methylation-based classification system for central nervous system (CNS) tumours has been developed, offering a more accurate and standardized diagnostic approach. This method, validated across a large reference cohort of 2,801 samples, enables precise classification of tumours into 82 distinct DNA methylation classes, with 52 of these corresponding to WHO-defined entities. The system was tested on 1,155 diagnostic cases, revealing that 88% of samples matched established methylation classes, with 12% of cases requiring re-evaluation due to diagnostic discordance. These re-evaluations led to revised diagnoses, including the reclassification of several tumours, such as IDH-wildtype astrocytomas as glioblastomas. The system demonstrated high accuracy, with a 99% area under the receiver operating characteristic curve (AUC), and a cross-validated error rate of 1.14% for clinically relevant groupings. A free online classifier tool is available for routine diagnostic use, allowing automated classification without additional onsite data processing. The method also provides a blueprint for developing machine learning-based tumour classifiers across other cancer types. Technical robustness was confirmed through inter-laboratory testing and compatibility with various DNA methylation platforms. The system has been implemented in five external centres, with 12% of cases showing new diagnoses. The approach offers significant clinical impact, improving diagnostic precision and potentially transforming tumour pathology. The study highlights the potential of DNA methylation profiling to enhance diagnostic accuracy and standardization in CNS tumour classification.A DNA methylation-based classification system for central nervous system (CNS) tumours has been developed, offering a more accurate and standardized diagnostic approach. This method, validated across a large reference cohort of 2,801 samples, enables precise classification of tumours into 82 distinct DNA methylation classes, with 52 of these corresponding to WHO-defined entities. The system was tested on 1,155 diagnostic cases, revealing that 88% of samples matched established methylation classes, with 12% of cases requiring re-evaluation due to diagnostic discordance. These re-evaluations led to revised diagnoses, including the reclassification of several tumours, such as IDH-wildtype astrocytomas as glioblastomas. The system demonstrated high accuracy, with a 99% area under the receiver operating characteristic curve (AUC), and a cross-validated error rate of 1.14% for clinically relevant groupings. A free online classifier tool is available for routine diagnostic use, allowing automated classification without additional onsite data processing. The method also provides a blueprint for developing machine learning-based tumour classifiers across other cancer types. Technical robustness was confirmed through inter-laboratory testing and compatibility with various DNA methylation platforms. The system has been implemented in five external centres, with 12% of cases showing new diagnoses. The approach offers significant clinical impact, improving diagnostic precision and potentially transforming tumour pathology. The study highlights the potential of DNA methylation profiling to enhance diagnostic accuracy and standardization in CNS tumour classification.