The study investigates the frontal midline theta (fmθ) activity during visual working memory tasks using EEG data from 23 subjects. The authors use independent component analysis (ICA) to decompose the EEG data into maximally independent components, revealing multiple spectral modes involving at least three frequency bands and a wide range of amplitudes. The fmθ components, which are primarily generated in or near the dorsal anterior cingulate cortex, exhibit significant increases in power during working memory tasks, particularly during the memorization of letters. These increases are not uniform across trials, with a small proportion of trials showing stronger theta activity. The study also finds that fmθ components produce 12–15 Hz low-beta activity during letter memorization, which is stronger during memorized trials compared to ignored trials. A new decomposition method, log spectral ICA, is introduced to analyze the time/frequency transforms of the fmθ component time courses, revealing harmonic energy in continuous, sharp-peaked theta wave trains and independent low-beta bursts. The results suggest that the variability in fmθ activity may reflect dynamic adjustments in medial frontal cortex to trial-specific behavioral context and task demands.The study investigates the frontal midline theta (fmθ) activity during visual working memory tasks using EEG data from 23 subjects. The authors use independent component analysis (ICA) to decompose the EEG data into maximally independent components, revealing multiple spectral modes involving at least three frequency bands and a wide range of amplitudes. The fmθ components, which are primarily generated in or near the dorsal anterior cingulate cortex, exhibit significant increases in power during working memory tasks, particularly during the memorization of letters. These increases are not uniform across trials, with a small proportion of trials showing stronger theta activity. The study also finds that fmθ components produce 12–15 Hz low-beta activity during letter memorization, which is stronger during memorized trials compared to ignored trials. A new decomposition method, log spectral ICA, is introduced to analyze the time/frequency transforms of the fmθ component time courses, revealing harmonic energy in continuous, sharp-peaked theta wave trains and independent low-beta bursts. The results suggest that the variability in fmθ activity may reflect dynamic adjustments in medial frontal cortex to trial-specific behavioral context and task demands.