Abstract Objective To investigate the differential ability of the “Test Relaties Abstracte Concepten” (TRACE), a Dutch test for abstract semantic knowledge, in frontotemporal dementia (FTD). Methods The TRACE was administered in patients with behavioral variant FTD (bvFTD; n = 16), nonfluent variant (nfvPPA; n = 10), logopenic variant (lvPPA; n = 10), and semantic variant primary progressive aphasia (svPPA; n = 9), and controls (n = 59). We examined group differences, performed correlational analyses with other neuropsychological tests and investigated discriminative ability. We compared the TRACE with a semantic association test for concrete stimuli (SAT). Results All patient groups, except nfvPPA, performed worse on the TRACE than controls (p < .01). svPPA patients performed worse than the other patient groups (p < .05). The TRACE discriminated well between patient groups, except nfvPPA, versus controls (all p < .01) and between svPPA versus other patient groups with high sensitivity (75–100%) and specificity (86%–92%). In bvFTD and nfvPPA the TRACE correlated with language tests (ρ > 0.6), whereas in svPPA the concrete task correlated (ρ ≥ 0.75) with language tests. Patients with bvFTD, nfvPPA and lvPPA performed lower on the TRACE than the SAT (p < .05), whereas patients with svPPA were equally impaired on both tasks (p = .2). Discussion We demonstrated impaired abstract semantic knowledge in patients with bvFTD, lvPPA, and svPPA, but not nfvPPA, with svPPA patients performing worse than the other subtypes. The TRACE was a good classifier between each patient group versus controls and between svPPA versus other patient groups. This highlights the value of incorporating semantic tests with abstract stimuli into standard neuropsychological assessment for early differential diagnosis of FTD subtypes.