The ability to maintain information for a short period of time (i.e. working memory, WM) tends to decrease across the life span with large inter-individual variability; the underlying neuronal bases, however, remain unclear. To address this issue, we used a multimodal imaging approach (voxel-based morphometry, diffusion-tensor imaging, electroencephalography) to test the contribution of brain structures and neural oscillations in an elderly population. Thirty-one healthy elderly participants performed a change-detection task with different load conditions. As expected, accuracy decreased with increasing WM load, reflected by power modulations in the theta-alpha band (5-12 Hz). Importantly, these power changes were directly related to the tract strength between parahippocampus and parietal cortex. Furthermore, between-subject variance in gray matter volume of the parahippocampus and dorsal striatum predicted WM accuracy. Together, our findings provide new evidence that WM performance critically depends on parahippocampal and striatal integrity, while theta-alpha oscillations may provide a mechanism to bind the nodes within the WM network.