The only head-to-head test of monaural against binaural beats, in fifteen epilepsy patients
- Year
- 2018
- Stimulus
- Theta
- Design
- Within-subject crossover with a beatless control, order randomized and counterbalanced
- Sample
- n = 15
Summary
Apps and forum posts argue about whether binaural or monaural beats work better, and the honest state of the evidence is that one study has compared them on something a person would notice. This is that study. It found memory performance ordered binaural, then control, then monaural, which is the result the argument usually assumes, and it found it in fifteen epilepsy patients who had electrodes inside their heads for surgical planning.
The design takes advantage of a rare opportunity. Fifteen patients being assessed before epilepsy surgery already had depth electrodes in the hippocampus and rhinal cortex, so the authors could record from inside the medial temporal lobe while the patients worked. Each patient did an associative learning task under three conditions in a randomized, counterbalanced order: 5 Hz monaural beats, 5 Hz binaural beats, and a control tone that was a plain 220 Hz sine wave with no beat in it. The beats themselves were built from 217.5 and 222.5 Hz tones. The task was scored two ways, for item memory (did you see this before) and source memory (which context did you see it in), and the authors recorded intracranial EEG throughout.
One detail of the procedure matters more than it first appears. The audio was not playing continuously in the background. It came on at the start of each trial and lasted only as long as that trial, 3.5 seconds during encoding and up to 5 seconds during retrieval. So this is not a study of listening to beats for a while and then being tested; it is a study of a few seconds of beats arriving at the moment a memory is formed or retrieved. That is a cleaner experiment than continuous exposure, and it is further from how anybody uses these recordings.
The behavioural result is a trend rather than a set of clean contrasts, and the distinction matters. Across the whole design there was a main effect of stimulation (F(2,24) = 4.45, p = 0.03) and a significant linear effect in the predicted order, binaural above control above monaural (F(1,12) = 5.59, p = 0.036). Break that into individual comparisons and most of them do not reach significance. For item memory, binaural beat monaural at p = 0.056 and the control at p = 0.085, both above the conventional threshold, with effect sizes of 0.27 and 0.13. For source memory three contrasts do land: binaural above monaural (p = 0.024, g = 0.47), control above monaural (p = 0.015, g = 0.34), while binaural above control does not (p = 0.162, g = 0.15). Read together, what carries the result is monaural beats being worse than everything else, more than binaural beats being better than the control.
On the EEG side both monaural and binaural stimulation increased phase locking of 5 Hz oscillations within rhinal cortex, and the authors’ key observation is that the two increases corresponded to reverse phase shifts. That is their proposed mechanism for the behavioural split: the same rhythm imposed with opposite timing relative to the brain’s own activity, helping encoding in one case and interfering with it in the other. It is a coherent story and this paper does not test it against alternatives.
We grade this mixed rather than by its score, which is 3.5. The findings are partial in the rubric’s sense: a significant overall trend whose component comparisons mostly miss, carried by one of the two memory measures, and the published memory literature around it disagrees with itself. Our other graded studies on beats and memory run from mixed through weak to unsupported, and nothing has pooled them.
Limitations. Fifteen patients, thirteen in the behavioural analysis, and the authors say “pilot study” in their own title. People with drug-resistant epilepsy, on anti-epileptic medication, with electrodes implanted in the structures being measured, are not a general population, and medial temporal lobe function is the reason they are in the study at all. The exposure is a few seconds per trial inside one laboratory session, and the outcome is a recognition task, so nothing here speaks to whether a track you play while studying does anything. No blinding is claimed. There is a single control condition and no silence arm. The effect sizes on the comparisons that reached significance, g = 0.47 and g = 0.34, are modest, and the ones that did not are smaller still. Being the only study of its kind makes this paper worth reading and does not make it strong.
How we graded this study
Each criterion is scored 1–5. Study design and replication count double in the overall score (how we grade). Graded from: full text read.
Graded by our research agent on Nov 16, 2026.