Our ability to maintain and later retrieve detailed visual memories is an everyday experience, and completing almost any task requires us to shuttle this information between working and long-term memory. However, the mechanisms by which the brain supports detailed, online memory representations and the temporal dynamics of these representations remain poorly understood. Another open question is the degree to which these representations evolve from short to long delays. Across three experiments, we tested how and when the brain represents feature specific memories –such as the color or location of an object– during both working memory and retrieval from long-term memory. These experiments employed a combination of continuous report measures and modeling of multivariate patterns of electroencephalography (EEG) activity to identify and characterize the patterns of oscillatory activity that encode the online maintenance of feature-specific memory representations. In Experiment 1, we explored which frequencies of EEG oscillations support online spatial working memories, and then use this activity to test the long-standing question of whether or not multiple items can be simultaneously maintained in working memory. In Experiment 2, we leveraged continuous report measures to examine how retrieval practice affects the probability that precise color memories can be retrieved and the precision with which that information can be retrieved from long-term memory. Finally, in Experiment 3, we combine these two approaches in order to test whether EEG activity can be used to obtain a time-resolved measure of long-term memory retrieval and to explore the degree to which patterns of rhythmic EEG activity at encoding are reinstated at retrieval. Taken together these experiments highlight the broad potential for using continuous report measures, EEG, and multivariate analysis to characterize memory function.