Oscillatory neural activity is a fundamental characteristic of the mammalian brain spanning multiple levels of spatial and temporal scale. Current theories of neural oscillations and analysis techniques employed to investigate their functional significance are based on an often implicit assumption: In the absence of experimental manipulation, the spectral content of any given EEG- or MEG-recorded neural oscillator remains approximately stationary over the course of a typical experimental session (∼1 h), spontaneously fluctuating only around its dominant frequency. Here, we examined this assumption for ongoing neural oscillations in the alpha-band (8:13 Hz). We found that alpha peak frequency systematically decreased over time, while alpha-power increased. Intriguingly, these systematic changes showed partial independence of each other: Statistical source separation (independent component analysis) revealed that while some alpha components displayed concomitant power increases and peak frequency decreases, other components showed either unique power increases or frequency decreases. Interestingly, we also found these components to differ in frequency. Components that showed mixed frequency/power changes oscillated primarily in the lower alpha-band (∼8–10 Hz), while components with unique changes oscillated primarily in the higher alpha-band (∼9–13 Hz). Our findings provide novel clues on the time-varying intrinsic properties of large-scale neural networks as measured by M/EEG, with implications for the analysis and interpretation of studies that aim at identifying functionally relevant oscillatory networks or at driving them through external stimulation.