(a) Schematic of experiment design. Two types of binocular rivalry stimuli consisting of image pairs (A,B) and (A,C), respectively, were presented. During one image pair block, 12 trials corresponding to the 12 positions of the two fixation spots were presented in randomized order before the next block corresponding to the other image pair was presented. This was repeated more than 60 times. (b) Decoding accuracy for distinguishing (A,B) from (A,C) was 74% (black vertical line), even though the conscious percept was A for both trial types. As a control, we shuffled labels 100 times and attempted to perform decoding. Gray bars show the distribution of decoding accuracies for these 100 shuffle iterations. (c) Scatterplot showing dominant stimulus modulation indices on the x-axis and suppressed stimulus modulation indices on the y-axis. Each triangle represents 1 of 66 physically selective cells recorded in one session from face patch ML with a 64-ch S-probe. (d) Schematic of three possible models for how perceptually modulated neurons may encode consciously perceived and suppressed stimuli during binocular rivalry. Left: (I) Neural responses encode the conscious percept in binocular rivalry identically to the corresponding unambiguous physical stimulus; x and y axes represent two dimensions of neural state space. Middle: (II) Responses during binocular rivalry lie in between the two stimuli but are biased toward the dominant stimulus. Right: (IIa) Spikes reflect a weighted sum of consciously perceived and suppressed stimuli and are generated through a Poisson process based on average firing rates. (IIb) Two different types of spikes, defined, for example through a temporal code, encode the consciously perceived and veridical physical stimulus, respectively. The time course in this schematic is from a single perceptual dominance period and divided into different epochs that represent either the conscious or physical stimulus.