Leptomeningeal metastases (LMs) are common metastatic complications, occurring in at least 5% of patients with disseminated cancer. Cerebrospinal uid (CSF) cytology remains the standard for diagnosis and assessment of treatment response, but may be inadequate. Our objective was to compare ventricular and lumbar CSF cytology in patients who had cytologically proven LM and were receiving intra-CSF chemotherapy. Sixty patients with LM, positive lumbar CSF cytology documented at diagnosis, limited extent of CNS disease, and no evidence of CSF ow obstruction were treated with a variety of intra-CSF chemotherapies. All patients underwent a single simultaneous ventricular and lumbar CSF sampling (mean volume of CSF per site examined, 10 ml) to assess response to therapy at either 1 or 2 months after treatment initiation. Ventricular CSF cytology was positive in 44 patients (73%), 35 of whom were also positive by lumbar CSF cytology. Lumbar CSF cytology was positive in 45 patients (75%), of which 35 were also positive by ventricular CSF cytology. Samples were negative at both ventricular and lumbar sites in 6 patients (10%). Paired CSF cytologies were discordant in 19 (32%) patients. The lumbar cytology was negative in 9, whereas the ventricular cytology was positive (lumbar false-negative rate of 17%); the ventricular cytology was negative in 10, whereas the lumbar cytology was positive (ventricular false-negative rate of 20%). In the presence of spinal signs or symptoms of LM, the lumbar CSF cytology was more likely to be positive than was the ventricular (odds ratio = 2.86; 95% con dence interval, 0.86-9.56). Conversely, in the presence of cranial signs or symptoms, the ventricular CSF cytology was more likely to be positive than was the lumbar (odds ratio = 2.71; 95% con dence interval, 0.76-9.71). In this cohort of patients, whose LM was documented initially by positive lumbar CSF cytology, ventricular and lumbar CSF samples obtained during treatment had similar false-negative rates, depending on the site of clinical or radiologic disease. This suggests that both lumbar and ventricular sites must be sampled when assessing treatment response. If clinical or radiographic disease is present only at 1 site, then CSF from that site is more likely to be positive than is CSF obtained from the more distant site. Neuro-Oncology 3, 42-45, 2001 (Posted to Neuro-Oncology [serial online], Doc. 00-033, October 19, 2000