We studied 210 participants (20-79 years old). Data from structural T1, diffusion tensor imaging, FLAIR, and resting state functional MRI scans combined with a neuropsychological evaluation were used. Z-scores for four cognitive domains measuring memory, fluid reasoning, speed of processing, language, and general cognition were computed. Brain volumes, thicknesses, fractional anisotropy tracts, functional connectivity, white-matter hyperintensity, age, educational and occupational level were used as the predictors, while cognitive performance served as the to-be-predicted outcome in linear multivariate regression models. We initially performed Principal Components Analysis (PCA) to capture the major sources of variance in each modality. By splitting the sample into training and test data sets, a PCA/Subprofile Scaling Model was then used for deriving a model from Principal-Component scores to fit cognitive performance in the training data set, with a subsequent prediction of cognitive performance based on derived model in the held-out test set. Analyses performed in the total sample, and in two age groups (young, old).