In the past several decades, experimentalists have proposed a large number of correlations to estimate steam condensation rates in the presence of noncondensable gases in free convection regimes. These correlations are largely empirical, and usually of limited scope, which often leads to their use outside their range of validity, thus incurring the risk of significant errors. In this investigation, we disregard the correlations altogether, and instead go back to their underlying original data, consolidate them in a single set, and propose a generalized correlation that is compatible with the heat and mass transfer analogy. This best-estimate correlation for steam-air mixtures, based on six different investigations and 350 data points, covers two orders of magnitude in the heat transfer coefficient, and is valid for pressures up to 20 bars and steam mass fraction from 0.1 to 0.95. The consolidated raw data are self-consistent and collapse around a curve with a standard deviation of 16%, thus well within typical experimental error bands. This demonstrates that there is no physical justification for the large deviations (factor 2 or more) observed sometimes when one compares the heat transfer coefficients predicted by different empirical correlations. Finally, the generalized correlation also shows that in the special case when the noncondensable gas density is held constant as the steam content is varied, the heat transfer coefficient can be expressed solely in terms of the steam to noncondensable gas density ratios. (C) 2015 Elsevier Ltd. All rights reserved.