The workings and products of the imagination are often described in visual terms: We speak of ‘mental images’ and the ‘mind’s eye.’ To what extent is this metaphorical? Should imagination be conceived of as a process of quasi-perceptual simulation or is it more sparse and abstract? Building on recent findings that suggest mental imagery tends to lack detail, we investigate how complete and coherent people’s imagined scenes are. In Experiment 1, we presented a riddle-like vignette to participants and found that they on average only imagined 54% of the simple features we asked them about. Moreover, successfully finding a solution was unrelated to the number of features imagined. In Experiments 2a – 2c we found that participants often did not notice spatial contradictions in text descriptions (2a), even when scaffolded with a map (2b), and that spotting these contradictions was unrelated to performance on a mental rotation task (2c).