A hallmark of working memory is its capacity limitation: memory gets poorer when the number of simultaneously memorized items increases. The neuronal origins of this limitation remain unclear. The capacity limitation is typically assessed by asking participants to report a single item out of several concurrently stored ones. Here, we applied a whole-report paradigm where participants reported all items. Specifically, participants memorized one, two or four orientations of Gabor gratings. After a delay, they were asked to rate their confidence about the memory quality (scale from 0=forgotten to 3=best possible remembered) and to reproduce the memorized orientation on a continuous scale, for every item. This allowed us to assess the number of items rated as forgotten and to estimate the aggregated memory precision for each trial. Participants performed 1600 trials in total while their brain activity was recorded with magnetoencephalography (MEG). Using MEG-beamforming, we applied multivariate pattern analysis to brain activity across the human cortical surface. Preliminary results suggest that activity patterns from posterior cortical regions predicted whether a subject rated items as forgotten. This prediction was pronounced during memory encoding but remained present until report. We found that the activity patterns in posterior regions also predicted the aggregated precision of the reproduced orientations. Interestingly, the precision-related signal resembled the temporal profile of the forgetting-related signal, indicating a common origin of capacity limitations in working memory. Future analyses will examine which factors contribute to the prediction signals including fluctuations of attention between trials or spatial positions and stimulation characteristics.