Working memory (WM) enables a rapid access to a limited number of items that are no longer physically present. WM studies usually involve the encoding and retention of multiple items, while probing a single item only. Hence, little is known about how well multiple items can be reported from WM. Here we asked participants to successively report each of up to 8 encoded Gabor patches from WM. Recall order was externally cued, and stimulus orientations had to be reproduced on a continuous dimension. Participants were able to sequentially report items from WM with an above-chance precision even at high set sizes. It is important that we observed that precision varied systematically with report order: It dropped steeply from the first to the second report but decreased only slightly thereafter. The observed trajectory of precision decrease across reports was better captured as a discontinuous rather than an exponential function, suggesting that items were reported from different states in visual WM. The following 3 experiments replicated these findings. In particular, they showed that the observed drop could not be explained by a retro-cueing benefit of the first report, a longer delay duration for later reports or a visual interference effect of the first report. Instead, executive interference of the first report reduced precision of subsequent reports. Together, the results show that a sequential whole-report procedure allows the assessment of qualitatively different states in visual WM.