The well-documented sex difference in mental rotation favoring males has been shown to emerge only for 2-dimensional presentations of 3-dimensional objects, but not with actual 3-dimensional objects or with virtual reality presentations of 3-dimensional objects. Training studies using computer games with mental rotation-related content have demonstrated training effects on mental rotation performance. Here, we studied the combined effect of a two-week mental rotation (MR) training on 2-dimensional vs. 3-dimensional presentations of a classic Shepard-Metzler task (presented in a pretest-training-posttest design) and their accompanying cortical activation patterns assessed via EEG in a sample of 38 male and 39 female adolescents of about 15 years of age. Analysis of one performance parameter (reaction times) displayed only main effects of dimensionality (with shorter RTs on the 3D vs. 2D version of the MR task) and of training (significant shortening of RTs), but no significant sex difference. Analysis of the other performance parameter (scores) in the MR task revealed a sex difference favoring males that first, appeared only in the 2D version, but not in the 3D version of the MR task and, secondly, diminished after training. Neurophysiologically we observed a complex sex × dimensionality × training × hemisphere interaction showing that the hypothesized decrease of brain activation (increase in neural efficiency) with training emerged for males in both 2D and 3D conditions, whereas for females this decrease was found only in the 3D but not with the 2D version of the MR task.