Phase contrast magnetic resonance imaging allows studying quantitatively the perimedullary cerebrospinal fluid (CSF) dynamics. However, the anatomy of the subarachnoid space difficults the segmentation of CSF due to the presence of vascular structures and spinal nerves. The aim of this paper is to describe a semiautomatic segmentation method for the study of the perimedullary CSF dynamics. The process is started with a seed point within the region to analyze. The algorithm creates a correlation map, calculates a threshold value and classifies pixels of CSF combining different temporal characteristics of flow behavior as input attributes to a k-means algorithm. One observer carried out ten times the segmentation of the cervical images in 5 healthy subjects; stroke volume and area were calculated. The variability of the obtained measurements was evaluated as an estimation of the reproducibility of the method. For this the coefficient of variation was calculated. The variability of the measurements was less than 5%. The method facilitates the quantification of perimedullary CSF. Stroke volume and the area at C2C3 space and prepontine cistern were measured in 16 healthy subjects.