Background: According to the personalization effect in multimedia learning, the use of personal and possessive pronouns in instructional materials (e.g., 'you' and 'your') is beneficial. However, current research suggests that the personalization effect is inverted for emotionally aversive content (e.g., illnesses). Objective: This study investigates whether a beneficial personalization effect can be observed for emotionally neutral health-related content whereas the effect may be reversed for emotionally aversive health-related content. Methods: In this study, 139 university students learned both emotionally aversive learning content on type 1 diabetes (within-factor) that was presented in either personalized or non-personalized language (between-factor). The presentation order of the content (neutral first vs. aversive first) was controlled (between-factor), resulting in a 2 × 2 × 2 mixed design. The dependent variables measured include learning outcomes (regarding retention and transfer), state anxiety, extraneous cognitive load, motivation and learning time. Results and Conclusions: In the transfer test, learners generally performed better when learning with non-personalized instructional materials than with personalized instructional materials, regardless of whether the content was emotionally neutral or aversive. The results raise questions regarding the robustness of the personalization effect and the underlying mechanisms of the inverted personalization effect. An alternative explanation to be investigated is whether the direct reference to a disease that the participants do not have (here: 'your type 1 diabetes') leads to schema interference, which could be responsible for poorer learning performance--even if the learning content about the disease can be considered emotionally neutral.