Complex human cognition, such as decision-making under ambiguity, is reflected in dynamic spatio-temporal activity in the brain. Here, we combined event-related potentials (ERPs), with their high temporal resolution and moderate spatial resolution afforded by high-density arrays, with computational modelling of the time course of decision-making and outcome evaluation. We contrasted four computational models (Expectancy-Valence, ProspectValence Learning (PVL), PVL-Delta, and Value Plus Perseverance (VPP) model) and found that the VPP model provided the best post hoc fit. Measures of choice probability generated using the VPP model, as well as objective trial outcomes, were applied as regressors in a general linear model of the EEG signal to create a three-dimensional spatio-temporal characterization of task-related neural dynamics. We observed that outcome valence, outcome magnitude, and VPP choice probability are expressed in distinctly separate components of the ERP. Our findings show, for the first time, model-based analysis of the spatio-temporal dynamics of outcome evaluation in complex human decision-making.