Degenerative alterations of cortical and subcortical regions in Alzheimer disease (AD) can be estimated by the extent of brain metabolite changes as measured by magnetic resonance spectroscopic imaging (MRSI). A neuropsychological assessment may correlate with metabolite levels and could evaluate underlying degenerative processes. Probabilistic-related classification learning, which represents one form of procedural learning, is associated with the neostriatum. The present study was aimed at examining the correlation of spectroscopic imaging in subcortical regions with the evaluation of specific neuropsychological findings. Twenty-two patients with Alzheimer's disease were compared to 15 healthy elderly control subjects. Proton MRSI of the basal ganglia (BG) and thalamus region was performed for detection of N-acetylaspartate (NAA), trimethylamine (TMA) and creatine ((P)Cr). In addition, a probabilistic-related classification learning task (Weather Prediction Task (WT)) was applied. We observed that in patients a high TMA signal in the basal ganglia region was correlated with a poorer performance in the probabilistic learning task (Spearman rank order correlation (SROC)=-0.6, P<0.009). Although Alzheimer's patients, as a group, did not differ from controls with regard to probabilistic learning capacity (PLC), male AD patients, as compared to male controls, displayed an impairment in the task performance by 28% (P<0.03) and showed a 16% elevation in TMA signaling (P<0.04). The altered metabolite signals and ratios in combination with the cognitive performance might suggest gender-related neuronal degeneration and dysfunction within subcortical regions in AD.