Oxygen isotopes in tree rings (δ18OTR) are widely used to reconstruct past climates. However, the complexity of climatic and biological processes controlling isotopic fractionation is not yet fully understood. Here, we use the MAIDENiso model to decipher the variability in δ18OTR of two temperature-sensitive species of relevant palaeoclimatological interest (Picea mariana and Nothofagus pumilio) and growing at cold high latitudes in North and South America. In this first modelling study on δ18OTR values in both northeastern Canada (53.86° N) and western Argentina (41.10° S), we specifically aim at (1) evaluating the predictive skill of MAIDENiso to simulate δ18OTR values, (2) identifying the physical processes controlling δ18OTR by mechanistic modelling and (3) defining the origin of the temperature signal recorded in the two species. Although the linear regression models used here to predict daily δ18O of precipitation (δ18OP) may need to be improved in the future, the resulting daily δ18OP values adequately reproduce observed (from weather stations) and simulated (by global circulation model) δ18OP series. The δ18OTR values of the two species are correctly simulated using the δ18OP estimation as MAIDENiso input, although some offset in mean δ18OTR levels is observed for the South American site. For both species, the variability in δ18OTR series is primarily linked to the effect of temperature on isotopic enrichment of the leaf water. We show that MAIDENiso is a powerful tool for investigating isotopic fractionation processes but that the lack of a denser isotope-enabled monitoring network recording oxygen fractionation in the soil–vegetation–atmosphere compartments limits our capacity to decipher the processes at play. This study proves that the eco-physiological modelling of δ18OTR values is necessary to interpret the recorded climate signal more reliably.