Background: Frontotemporal dementia (FTD) is a heterogeneous disease both at the clinical, genetic, and pathobiological level. The underlying pathological spectrum (termed FTLD, frontotemporal lobar degeneration) is in most cases defined by accumulation of either tau (FTLD-tau) or TDP-43 proteins (FTLD-TDP). Biomarkers to differentiate these subtypes are not yet available, whereas these are essential requirements to study the natural course of disease and for homogeneous inclusion of patients in clinical studies. Objective: To study if a combination of total (t-) and phosphorylated (p-)tau, and t-TDP-43 and p-TDP-43 proteins in cerebrospinal fluid (CSF) is suitable to discriminate FTLD-tau and FTLD-TDP subtypes. Methods: We developed immunoassays for the quantification of t-TDP-43 and p-TDP-43 proteins and used commercially available assays for the quantification of t-tau and p-tau proteins. We quantified these proteins in ventricular CSF samples from neuropathologically defined FTLD-tau and FTLD-TDP cases to study the reflection of underlying brain pathology in CSF composition, and in lumbar CSF samples from FTLD-tau and FTLD-TDP patients to study the diagnostic potential of CSF biomarkers. Results: In ventricular CSF, t-TDP-43 and t-tau levels, when combined into one model, were significantly different between neuropathologically-defined FTLD-tau and FTLD-TDP cases. In a pilot study using lumbar CSF, the p-tau/t-tau ratio, but not t-TDP-43 levels, were significantly different between FTLD-TDP and FTLD-tau patients. Conclusion: We conclude that with current available methods, CSF tau, rather than TDP-43 proteins, may have diagnostic value in the differentiation of FTLD patients with either tau or TDP-43 pathology.