A knowledge-based fuzzy clustering (KBFC) MRI segmentation algorithm was proposed to obtain accurate tumor segmentation for tumor volume measurement of nasopharyngeal carcinoma (NPC). An initial segmentation was performed on T1 and contrast enhanced T1 MR images using a semi-supervised fuzzy c-means (SFCM) algorithm. Then, three types of anatomic and space knowledge--symmetry, connectivity and cluster center were used for image analysis which contributed the final tumor segmentation. After the segmentation, tumor volume was obtained by multi-planimetry method. Visual and quantitative validations were performed on phantom model and six data volumes of NPC patients, compared with ground truth (GT) and the results acquired using seeds growing (SG) for tumor segmentation. In visual format, KBFC showed better tumor segmentation image than SG. In quantitative segmentation quality estimation, on phantom model, the matching percent (MP) / correspondence ratio (CR) was 94.1-96.4% / 0.888-0.925 for KBFC and 94.1-96.0% / 0.884-0.918 for SG while on patient data volumes, it was 92.1+/- 2.6% / 0.884+/- 0.014 for KBFC and 87.4+/- 4.3% / 0.843+/- 0.041 for SG. In tumor volume measurement, on phantom model, measurement error was 4.2-5.0% for KBFC and 4.8-6.1% for SG while on patient data volumes, it was 6.6+/- 3.5% for KBFC and 8.8+/- 5.4% for SG. Based on these results, KBFC could provide high quality of MRI tumor segmentation for tumor volume measurement of NPC.