Purpose Here, we evaluate the sensitivity and specificity of a new 11-parameter flow cytometry (FCM) approach versus conventional cytology (CC) for detecting neoplastic cells in stabilized CSF samples from newly diagnosed aggressive B-cell non-Hodgkin's lymphoma (B-NHL) at high risk of CNS relapse, using a prospective, multicentric study design. Patients and Methods Moreover, we compared the distribution of different subpopulations of CSF leukocytes and the clinico-biologic characteristics of CSF+ versus CSF−, patients, in an attempt to define new algorithms useful for predicting CNS disease. Results Overall, 27 (22%) of 123 patients showed infiltration by FCM, while CC was positive in only seven patients (6%), with three other cases being suspicious (2%). CC+/FCM+ samples typically had more than 20% neoplastic B cells and/or ≥ one neoplastic B cell/μL, while FCM+/CC− samples showed lower levels (P < .0001) of infiltration. Interestingly, in Burkitt lymphoma, presence of CNS disease by FCM could be predicted with a high specificity when increased serum β2-microglobulin and neurological symptoms coexisted, while peripheral blood involvement was the only independent parameter associated with CNS disease in diffuse large B-cell lymphoma, with low predictive value. Conclusion FCM significantly improves the sensitivity of CC for the identification of leptomeningeal disease in aggressive B-NHL at higher risk of CNS disease, particularly in paucicellular samples.