We report two experiments on the influence of spatial caricaturing on face learning and recognition. In the learning phase of Experiment 1, participants learned unfamiliar faces, half of which were caricatured at an exaggeration level of 30%. During learning we found increased N170, N250 and LPC and decreased P200 for caricatures. At test we observed better generalization to non-identical exemplars for faces learned as caricatures. At test, N250 was larger for caricatures. In Experiment 2, using more and stronger levels of caricaturing (0%=veridical, 35% and 70%), we replicated and extended main findings of Experiment 1 and found a clear learning advantage for caricatures. ERP and performance effects were directly related to the level of caricaturing. Overall, the present results suggest that (i) distinctive shape information is particularly informative to form new face representations for unfamiliar faces, and (ii) this formation of new representations is reflected by an increase in occipito-temporal negativity that is most prominent in the time range of the P200 and N250 components.