The objective of this study was to examine the diagnostic accuracy of imaging and CSF biomarkers in clinically ambiguous dementia (CAD). 69 patients were prospectively followed. The endpoint was clinical diagnosis at follow-up of 24 months based upon existing criteria. Medial temporal lobe atrophy score on MRI, distinctive patterns on 99 mTc-HMPAO-SPECT, and CSF levels of amyloid-β peptide, total tau protein, and P-tau181P were used together with neuropsychological testing to assess Se (sensitivity) and Sp (specificity) of separate and combined markers. 60 patients reached the endpoint. A definite diagnosis was achieved in 48 patients. CSF biomarkers had a Sp of 71% and a Se of 100% for Alzheimer's disease (AD) diagnosis. Sp increased to 88% and 93% when MRI and MRI + SPECT were combined, at the expense of Se. CSF biomarkers levels also provided clues to frontotemporal (FTD) or vascular dementias (VaD) diagnosis when situated in an intermediate range between normal and pathological values. MRI and SPECT contributed mostly to the diagnosis of VaD (Se 88%, Sp 75%) and FTD (Se 73%, Sp 78%), respectively. Initial neuropsychological testing had a poor diagnostic accuracy, except for a neuropsychiatric inventory score >40 for the diagnosis of FTD (Se 73%, Sp 84%). Independent of the clinical diagnosis, medial temporal lobe atrophy and total-tau were best correlated with cognitive decline at 2 years. In conclusion, CSF biomarkers efficiently predict evolution toward an AD phenotype in CAD. Imaging biomarkers mostly contribute to the differential diagnosis between non-AD dementias. Initial neuropsychological testing was poorly contributive in CAD diagnosis.