In multiple sclerosis, brain-reactive T cells invade the central nervous system (CNS) and induce a self-destructive inflammatory process. T-cell infiltrates are not only found within the parenchyma and the meninges, but also in the cerebrospinal fluid (CSF) that bathes the entire CNS tissue1,2. How the T cells reach the CSF, their functionality, and whether they traffic between the CSF and other CNS compartments remains hypothetical3,4,5,6. Here we show that effector T cells enter the CSF from the leptomeninges during Lewis rat experimental autoimmune encephalomyelitis (EAE), a model of multiple sclerosis. While moving through the three-dimensional leptomeningeal network of collagen fibres in a random Brownian walk, T cells were flushed from the surface by the flow of the CSF. The detached cells displayed significantly lower activation levels compared to T cells from the leptomeninges and CNS parenchyma. However, they did not represent a specialized non-pathogenic cellular sub-fraction, as their gene expression profile strongly resembled that of tissue-derived T cells and they fully retained their encephalitogenic potential. T-cell detachment from the leptomeninges was counteracted by integrins VLA-4 and LFA-1 binding to their respective ligands produced by resident macrophages. Chemokine signalling via CCR5/CXCR3 and antigenic stimulation of T cells in contact with the leptomeningeal macrophages enforced their adhesiveness. T cells floating in the CSF were able to reattach to the leptomeninges through steps reminiscent of vascular adhesion in CNS blood vessels, and invade the parenchyma. The molecular/cellular conditions for T-cell reattachment were the same as the requirements for detachment from the leptomeningeal milieu. Our data indicate that the leptomeninges represent a checkpoint at which activated T cells are licensed to enter the CNS parenchyma and non-activated T cells are preferentially released into the CSF, from where they can reach areas of antigen availability and tissue damage.