EEG records electrical potentials generated by synchronized neuronal populations across the scalp, with millisecond-level temporal precision. Event-related potentials (ERPs) isolate transient neural responses locked to specific events by averaging across trials, revealing components like the P300 (attention) and N400 (semantic processing) that index cognitive operations. ERP timing reveals the sequence of cognitive processes in a way that imaging methods cannot.
From your biology prerequisites, you know that neurons communicate via action potentials and that postsynaptic potentials sum to influence whether a cell fires. EEG capitalizes on a specific subset of this activity: when large populations of pyramidal neurons in cortical layers fire in synchrony, their combined postsynaptic potentials create electrical fields large enough to be detected at the scalp. Crucially, EEG does *not* primarily record action potentials (which are brief, poorly synchronized, and cancel out over large populations) — it records the slow, graded postsynaptic potentials of thousands of aligned neurons summing their activity. The resulting signal is a continuous voltage trace measured in microvolts, and because electrical fields propagate almost instantaneously, the temporal resolution is on the order of single milliseconds.
This temporal precision is EEG's defining advantage. Brain imaging methods like fMRI measure the BOLD signal — a hemodynamic response that unfolds over 4–6 seconds — and can only tell you *where* activity occurred, not *when* within the cognitive process. EEG inverts this: it has poor spatial resolution (the scalp-recorded signal is smeared by the skull and scalp) but superb temporal resolution. The fundamental trade-off between spatial and temporal resolution in neuroimaging means that EEG and fMRI are not competitors but complements — they answer different questions. If you want to know whether a semantic judgment occurs at 250 ms or 400 ms after a word appears, EEG is the tool. From your Fourier analysis prerequisite, you also know how to decompose a continuous signal into its frequency components — this is exactly how researchers analyze EEG frequency bands: delta (1–4 Hz, slow-wave sleep), theta (4–8 Hz, memory encoding), alpha (8–12 Hz, relaxed wakefulness, inhibition), beta (12–30 Hz, active cognition), and gamma (>30 Hz, local binding and feature integration).
Event-related potentials (ERPs) are extracted from the continuous EEG by averaging across many trials aligned to the same event (a tone, a word, a decision). The logic is pure signal-to-noise: the ERP component is time-locked to the event and will consistently appear at the same latency, while background EEG noise is random and will average toward zero across many trials. A typical ERP experiment averages 30–100+ trials per condition. The result is a waveform with labeled peaks and troughs defined by polarity (P = positive, N = negative) and latency in milliseconds. The P300 (a positive deflection peaking around 300 ms) is elicited by rare, task-relevant targets and indexes the allocation of attentional resources and working memory updating — it is larger when the target is more surprising and more attended. The N400 (a negative deflection peaking around 400 ms) is elicited by semantically unexpected words (e.g., "He spread butter on his *dog*") and reflects the ease or difficulty of lexical-semantic integration — a larger N400 means harder integration.
The power of ERP methodology lies in temporal sequencing: by identifying which components appear and when, researchers reconstruct the timeline of cognitive processing. For example, syntactic violations elicit an early left-anterior negativity (ELAN) around 100–200 ms, while semantic violations elicit the N400 around 400 ms — suggesting that syntactic analysis precedes semantic integration. This kind of millisecond-level resolution of cognitive sub-processes is entirely invisible to imaging methods, and it connects directly to the cognitive models you encounter in other courses: the sequence of ERP components provides empirical constraints on what happens first, what happens in parallel, and what depends on what.