This paper investigates model uncertainty in cross-country growth regressions using Bayesian Model Averaging (BMA). The authors find that the posterior probability is spread across many models, suggesting BMA is superior to choosing a single model. Out-of-sample predictive results support this. While Levine and Renelt (1992) found few significant regressors, Sala-i-Martin (1997b) found many important variables. The authors use BMA to address model uncertainty, allowing for any subset of up to 41 regressors. They use Markov chain Monte Carlo (MCMC) techniques to handle the large number of models (2.2 trillion). Their results broadly support Sala-i-Martin's more optimistic conclusion that some variables are important for growth. However, the variables identified differ slightly. The authors advocate BMA, where all inference is averaged over models using posterior probabilities. This approach provides a clear interpretation and formal statistical basis for inference and prediction. The paper also highlights the benefits of BMA over traditional model selection, showing it leads to better predictive performance. The authors conclude that BMA is recommended for growth regression due to its predictive advantage and formal statistical framework.This paper investigates model uncertainty in cross-country growth regressions using Bayesian Model Averaging (BMA). The authors find that the posterior probability is spread across many models, suggesting BMA is superior to choosing a single model. Out-of-sample predictive results support this. While Levine and Renelt (1992) found few significant regressors, Sala-i-Martin (1997b) found many important variables. The authors use BMA to address model uncertainty, allowing for any subset of up to 41 regressors. They use Markov chain Monte Carlo (MCMC) techniques to handle the large number of models (2.2 trillion). Their results broadly support Sala-i-Martin's more optimistic conclusion that some variables are important for growth. However, the variables identified differ slightly. The authors advocate BMA, where all inference is averaged over models using posterior probabilities. This approach provides a clear interpretation and formal statistical basis for inference and prediction. The paper also highlights the benefits of BMA over traditional model selection, showing it leads to better predictive performance. The authors conclude that BMA is recommended for growth regression due to its predictive advantage and formal statistical framework.