When we view an image, visual regions of the brain automatically respond to various properties of the image, including its category, size, and content. Recent work has shown that the visual information we freely recall from memories is highly detailed (Bainbridge et al., 2018). However, it is still largely unknown how content-based representations of stimuli during perception compare to those of free recall in the brain. In this study, participants (N=22) in a 7T MRI performed a free recall task in which they studied an image for 6s, followed by a 6s distractor task with separate images, and then were asked to freely recall the original image (6s) and then indicate vividness of their memory. In contrast to many prior studies, the recall did not require the learning of any stimulus associations and all stimuli were trial-unique. The 192 stimuli were organized in nested categorical structure: objects varying along two factors of big/small and tool/non-tool, and scenes varying along the factors of open/closed and natural/manmade. Each of these categories contained three object or scene types (e.g., guitar, living room), with eight exemplars of each type. Following the scan, participants were given a surprise recognition memory test. We find that key visual regions (e.g., the parahippocampal place area, lateral occipital cortex) contained information during encoding predictive of recognition and recall performance, while the hippocampus reflected recall vividness. Visual regions also showed categorical structure at all levels (object vs. scene; open vs. closed; guitar vs. cupcake) during both encoding and recall. Interestingly, while the hippocampus showed sensitivity to broad stimulus category during encoding, it did not show clear evidence for sub-categorical structure. In sum, we find preserved fine-grained categorical representations in visual regions during both encoding and recall, while the representational structure may change in more memory-related regions.