Humans and primates can quickly recognize mirror images of previously exposed pictures. This spontaneous mirror invariance, though advantageous for visual recognition, makes it difficult to distinguish the orientation of letters (e.g. to differentiate a “b” from a “d”), and may result in classical mirror reading and writing errors in preschoolers. Mirror invariance must therefore be overcome during reading acquisition. The Visual Word Form Area (VWFA), a region in the ventral stream that develops with reading expertise, was previously shown to discriminate words from their mirror images in literate adults. Here we investigate whether this region underlies mirror-image discrimination at the most elementary level of the orthographic code, the single-letter level. Using an fMRI priming paradigm, we demonstrate that the VWFA distinguishes the left–right orientation of single letters in skilled readers, and yet exhibits mirror invariance for simple pictures of matched complexity. These results clarify how letter shapes, after reading acquisition, escape the process of mirror invariance which is a basic property of the ventral visual shape recognition pathway.