Abstract
Discrete resource models of visual working memory (WM) suggest that the “currency” of WM is best
understood in terms of items rather than in terms of the number of independent features that are
stored. For instance, it is possible that WM storage is limited by a “pointer system” that can select a
maximum number of individuated representations, without regard to the specific features that are
maintained, or the total number of features associated with each item. Here, we show that ongoing
EEG activity, analyzed via multivariate classification of voltage topography, provides positive
evidence for feature-independent storage in visual WM. Using visual displays that controlled for the
total amount of sensory stimulation, we manipulated the number of items stored in visual WM (1-4),
while varying the feature content of the memoranda. A common group of subjects participated in 3
separate EEG sessions, during which memoranda varied between colors, orientations, and colored
oriented lines. Using logistic regression, we observed robust decoding of the number of targets in
the test display for each type of stimulus, even at the single-trial level. Critically, cross-training
analyses showed that a multivariate model trained on one feature (e.g., color) could effectively
decode the load for a different feature (e.g., orientation) as well as for conjunction stimuli that
contained double the number of relevant features. In the latter case, a single item with one relevant
feature (e.g., color or orientation) yielded the same load signature as an item with double the feature
load (color and orientation), indicating that this multivariate index of WM load is independent of the
total number of features stored. These findings highlight a time-resolved method for tracking storage
loads in visual WM, while providing positive evidence for a feature-independent aspect of storage
that may underlie item limits in visual working memory.