Somatic embryogenesis in gymnosperms is an effective approach to clonally propagating germplasm. However, embryogenic cultures frequently lose regenerative capacity. The interactions between metabolic composition, physiological state, genotype and embryogenic capacity in Pinus taeda (loblolly pine) somatic embryogenic cultures were explored using metabolomics. A stepwise modelling procedure, using the Bayesian information criterion, generated a 47 metabolite predictive model that could explain culture productivity. The model performed extremely well in cross-validation, achieving a correlation coefficient of 0.98 between actual and predicted mature embryo production. The metabolic composition and structure of the model implied that variation in culture regenerative capacity was closely linked to the physiological transition of cultures from the proliferation phase to the maturation phase of development. The propensity of cultures to advance into this transition appears to relate to nutrient uptake and allocation in vivo, and to be associated with the tolerance and response of cultures to stress, during the proliferation phase.