Multisensory learning to read involves, to a large extent, learning to write. A major problem in the initial teaching of handwriting is preventing children from producing reversed letters, especially when the reversed letters are identical to other letters. Torres et al. (2021) offer an efficient method for remediating this problem. Here, we analyze the reversals in their writing data, obtained on Brazilian first-graders (Mage = 6.0 years). Surprisingly, this analysis led to the observation that the first graders almost systematically reverse both the letters b and d in the particular copying conditions (the students look at one letter at a time for 3 s, then immediately after they had to write it while blindfolded). We first describe succinctly and discuss three models susceptible to account for reversal writing, with the aim to question their capacity of account for the curious observation just mentioned. The three models respectively attribute a major role to 1) initial (perceptive) mirror equivalence, 2) intra-hemispheric transfer, 3) orientation of the letters. Because none of the three models examined accounts convincingly for the observation, we accommodated and specified Model 2, adding also a major idea of Model 3. The resulting model assumes that the mirror-letter reversed image representation (b for d and vice-versa) is strongly activated in the right cerebral hemisphere, and that the top-down processes originating from this hemisphere were exacerbated by the eyes closed condition. Of course, this post-hoc and speculative model should be tested in other conditions and with other children.