In order to solve the problem of insufficient segmentation of brain tumors in magnetic resonance imaging (MRI) caused by noise, poor contrast, and diffused boundaries of tumors, a three-dimensional (3D) segmentation algorithm for brain tumor MRI images based on the improved continuous max-flow is proposed in this paper. Firstly, three types of MIR images, Flair, T1C and T2, are pre-processed with median filtering and fast fuzzy C means clustering. Then, the pre-processed images are linearly fused in the ratio of 5∶1∶4 (Flair, T1C, and T2) which is statistically observed from a large amount of experiments. Next, the 3D fused image is clustered by the fast fuzzy C-means algorithm to obtain the 3D under-segmented image. Finally, the proposed improved continuous max-flow algorithm acts on the 3D under-segmented image to obtain the final segmentation result with scattering points removed according to the analysis of the structural features and statistical characteristics of the 3D under-segmented image. The experimental results show that the average Dice coefficient, precision, and recall of the proposed method relative to the gold standard is up to 0.90, 0.94, and 0.86, respectively. The proposed algorithm can realize the 3D segmentation of the target regions precisely and automatically to meet the clinical medicine requirement.