Scalp voltage from millisecond-scale cortical synchrony — the cheapest, most portable, and most widely used window into human brain dynamics.
EEG electrodes on the scalp pick up voltage differences produced by synchronous postsynaptic currents in cortical pyramidal neurons, not the action potentials themselves. Pyramidal cells are arranged in near-parallel sheets, so thousands of them firing together sum to a field that escapes the skull. That synchrony is the signal: a single neuron is invisible, a coherent population is not.
The resulting waveforms are on the order of 10–100 µV. The dynamic range matters more than the amplitude: since the same range carries the signal, it also sets the noise floor that amplifier quality and referencing must beat.
The international 10–20 system places electrodes at 10% and 20% distances from nasion, inion, and the preauricular points, giving a reproducible whole-scalp montage. Sample rates of 250–2048 Hz are typical, with a practical passband near 0.1–40 Hz after filtering.
Referencing is a modelling decision, not a detail. EEG records a potential difference between sites, so every voltage is stated relative to something. A common average reference, a linked-mastoid reference, and a single-electrode reference all produce different-looking data from the same brain. Report the reference, and re-reference when comparing across montages.
Scalp EEG has no principled inverse solution. The skull is roughly 80× less conductive than the brain and the cerebrospinal fluid far more conductive, so the field smears spatially as it passes outward. Activity is blurred, deep sources are attenuated, and there is no unique set of sources producing a given scalp map.
Practical tools exist, none of which manufacture resolution that was never there. A surface Laplacian estimates the field's second derivative to sharpen topography. sLORETA and eLORETA normalise minimum-norm estimates for the expected variance of each voxel. Independent component analysis removes eye and muscle components as sources before averaging. The defensible conclusion from scalp EEG alone is a statistical map of relative synchrony — not a picture of the cortex.
Artefacts are large and mostly identifiable, which makes EEG unusually transparent once you know what to look for.
Averaging to the event, time-locked to a stimulus or response, cancels activity that varies in timing and leaves the response that does not. The result is the event-related potential — a component measured in microvolts, at a latency in milliseconds, relative to an explicit baseline.
The standard inventory: P1/N1 (early sensory), N170 (face-selective processing, occipitotemporal, ~170 ms), MMN (automatic deviance detection, ~150–250 ms), N2pc (target discrimination, ~200–300 ms), and P300 (context updating, ~300–500 ms). Latencies and amplitudes scale with stimulus evidence, so ERP effects map cleanly onto the drift rate v in a Drift Diffusion Model .
P300 amplitude is largest for rare, task-relevant, attended events — the three factors in the novelty P3 account. A large P300 to a frequent, ignored stimulus usually signals that the task was not attended to as designed, which is a finding about the experiment, not the brain.