Brain glucose metabolism as assessed by [18F]FDG positron emission tomography (PET) is expected to be significantly related to resting-state functional MRI (rs-fMRI) activity and functional connectivity (FC), but the underlying coupling model is still incompletely understood. Employing simultaneous acquisitions, we related [18F]FDG standard uptake value ratio (SUVR) to 50 features pertaining to rs-fMRI 1) signal, 2) hemodynamic response, 3) static and 4) time-varying FC, and 5) phase synchronization. To assess which rs-fMRI variables better describe SUVR across regions, we employed a hierarchical approach, identifying the model at population level, and then estimating it on individual data. Multilevel modelling explained around 40% of the SUVR variance, with signal-related features as the most relevant fMRI variables. When the model was used to characterize between-network variability of the SUVR-fMRI coupling, the ranking changed. We demonstrate that local activity and synchronization are the most important predictors of glucose metabolism, while large-scale FC properties gain importance within specific networks.