Attribute amnesia (Chen & Wyble, 2015, 2016) demonstrates that we may not always be able to spontaneously retrieve a simple attribute of a visual object (e.g., its color) for conscious report, even though the object had just been the target in a visual task. Attribute amnesia has been suggested to reflect a lack of consolidation of the task-irrelevant attribute in visual working memory. Here we tested whether saccadic selection eliminates or attenuates attribute amnesia. Saccade targets have been shown to be preferentially encoded into visual working memory and may therefore be spared. We used simple color pop-out displays, asking participants to indicate the location of the color singleton letter target on each trial either by keypress or by making a saccade toward it. After a couple of trials and unannounced to the participants, we asked for the color and identity of the last target letter on a surprise trial. We found that saccade targets were not spared from attribute amnesia: Participants were as bad in correctly reporting the color in the saccade as in the keypress condition. For letter identity, the effect was attenuated but not abolished when the target was foveated for a short period of time. We argue that the current results do not refute an obligatory coupling between saccadic selection and encoding in visual working memory. However, the encoded information may not necessarily be stored in a manner that is robust enough to persist in the face of a surprise question.