Phosphomannomutase-2 deficiency (PMM2-CDG) is the most frequent congenital disorder of glycosylation and it is associated with a recognizable facial pattern. There are no phenotype-genotype correlations neither early severity predictors. We evaluate dysmorphic features and propose a classification dividing them into major and minor with diagnostic implications.  We elaborated a severity dysmorphology categorization with proven predictive value. We trained a digital facial recognition platform (Face2Gene) that is now able to identify PMM2-CDG phenotypes at different ages. Both dysmorphology and digital analysis may help physicians to diagnose this chameleonic disease. Moreover, some early dysmorphic features are able to predict long-term clinical severity. (Dr. Toni Martinez Monseny, https://jmg.bmj.com/content/early/2018/11/21/jmedgenet-2018-105588 )