Little is known of combined utility of magnetic resonance imaging (MRI) and cerebrospinal fluid (CSF) biomarkers for prediction of Alzheimer's disease (AD) and longitudinal data is scarce. We examined these biomarkers at baseline and longitudinally in incipient AD. Forty-five subjects [21 controls (NL-NL), 16 stable MCI (MCI-MCI), 8 MCI who declined to AD (MCI-AD)] received MRI and lumbar puncture at baseline and after 2 years. CSF measures included total and phosphorylated tau (T-tau, P-tau(231)), amyloid-beta (Abeta(42)/Abeta(40)) and isoprostane. Voxel-based morphometry identified gray matter concentration (GMC) differences best distinguishing study groups and individual GMC values were calculated. Rate of medial temporal lobe (MTL) atrophy was examined using regional boundary shift (rBS) method. At baseline, for MRI, MCI-AD showed reduced GMC-MTL, and for CSF higher CSF T-tau, P-tau(231), IP and lower Abeta(42)/Abeta(40) as compared with MCI-MCI or NL-NL. Longitudinally, rBS-MTL atrophy was higher in MCI-AD than in either MCI-MCI or NL-NL, particularly in the left hemisphere. CSF data showed longitudinally greater increases of isoprostane in MCI-AD as compared with NL-NL. Combining baseline CSF-P-tau(231) and GMC-MTL significantly increased overall prediction of AD from 74% to 84% (p(step)<0.05). These results provide support for including multiple modalities of biomarkers in the identification of memory clinic patients at increased risk for dementia.