This paper examines the robustness of explanatory variables in cross-country economic growth regressions. It employs a novel approach, Bayesian Averaging of Classical Estimates (BACE), which constructs estimates as a weighted average of OLS estimates for every possible combination of included variables. The weights applied to individual regressions are justified on Bayesian grounds in a way similar to the well-known Schwarz criterion. Of 32 explanatory variables we find 11 to be robustly partially correlated with long-term growth and another five variables to be marginally related. Of all the variables considered, the strongest evidence is for the initial level of real GDP per capita.