20Perceptual decisions depend on coordinated patterns of neural activity cascading across 21 the brain, running in time from stimulus to response and in space from primary sensory 22 regions to the frontal lobe. Measuring this cascade and how it flows through the brain is 23 key to developing an understanding of how our brains function. However observing, let 24 alone understanding, this cascade, particularly in humans, is challenging. Here, we report 25 a significant methodological advance allowing this observation in humans at 26 unprecedented spatiotemporal resolution. We use a novel encoding model to link 27 simultaneously measured electroencephalography (EEG) and functional magnetic 28 resonance imaging (fMRI) signals to infer the high-resolution spatiotemporal brain 29 dynamics taking place during rapid visual perceptual decision-making. After 30 demonstrating the methodology replicates past results, we show that it uncovers a 31 previously unobserved sequential reactivation of a substantial fraction of the pre-response 32 network whose magnitude correlates with decision confidence. Our results illustrate that 33 a temporally coordinated and spatially distributed neural cascade underlies perceptual 34 decision-making, with our methodology illuminating complex brain dynamics that would 35 otherwise be unobservable using conventional fMRI or EEG separately. We expect this 36 methodology to be useful in observing brain dynamics in a wide range of other mental 37 processes. 38 39 40 Many previous studies have used known EEG markers (P1, N2, N170, P300, α-63 rhythm) or data driven approaches such as Independent Component Analysis (ICA) to 64 combine EEG with fMRI data 4,8-16 . One promising approach has been to use supervised 65 machine-learning techniques (e.g. classifiers) to find relevant projections of the EEG 66 data, where single-trial variability of the electrophysiological response along these 67 projections can be correlated in the fMRI space. Goldman, et al. 17 , Walz, et al. 18 and 68 Fouragnan, et al. 19 have demonstrated this technique on visual and auditory paradigms. 69This methodology has been shown to localize cortical regions that modulate with the task 70 while preserving the temporal progression of task-relevant neural activity. 71Here we combine a classification methodology with an encoding model that 72relates the trial-to-trial variability in the EEG to what is observed in the simultaneously 73 acquired fMRI. Encoding models have become an important machine learning tool for 74 analysis of neuroimaging data, specifically fMRI 20 . In most cases encoding models have 75 been used to learn brain activity that encodes or represents features of a stimulus, such as 76 visual orientation energy in an image/video 21-23 , acoustic spectral power in sound/speech 77 24 , or visual imagery during sleep 25 . In the method presented here, we employ an 78 encoding model to directly relate the simultaneously collected data from the two 79 neuroimaging modalities-instead of features derived from the st...