Background: Prediction models for decision-making in oncology are increasingly being used, but few are available for esophagogastric cancer, particularly in the metastatic setting. The aim of this study is to construct prediction models for overall survival in patients with metastatic esophageal or gastric cancer. Methods: Data from patients with metastatic esophageal (N = 8670) and gastric (N = 4804) cancer diagnosed in the period 2005-2015 were retrieved from the nationwide Dutch cancer registry. Multivariate Cox regression models, extended with treatment interactions, were created to predict overall survival. Multiple imputations were used to handle missing data. Predictor selection was performed via the Akaike Information Criterion (AIC) and was extended by a Delphi consensus among experts in the field of palliative esophagogastric cancer. Validation was performed with an 11-fold temporal validation. Both the concordance-index (c-index) and calibration were used to assess model quality. Results: The Delphi consensus yielded seven important predictors of survival and are shown with the AIC-selected predictors in Table 1. The c-indices show consistent discriminative ability during validation, i.e. 0.71 and 0.68 for respectively the esophageal and gastric cancer models. There is close agreement between predicted and observed survival, with an error of 1.7% and 2.2% for respectively the esophageal and gastric cancer models. Conclusions: The models yield fair discrimination and high calibration levels, and provide a good foundation for further investigation in clinical practice to determine their added value in decision-making.