Almost all available literature correlations to predict the volumetric gas-liquid mass transfer coefficient, k(L)a in agitated reactors are systems- or operating conditions-dependent. In this study, two back-propagation neural networks (BPNNs), one dimensional and one dimensionless were developed to correlate k(L)a for numerous gas-liquid systems in both surface-aeration reactors (SAR) and gas-inducing reactors (GIR) operating under wide ranges of industrial conditions. A total of 4435 experimental data points obtained from more than 10 publications for 50 gas-liquid systems were used to train, validate the dimensional and dimensionless BPNNs, which were able to correlate all k(L)a values with R-2 of 90.5 and 88.6%, respectively. The dimensional BPNN was used to predict the effect of various operating parameters on k(L)a in a number of important industrial processes. The predictions showed that increasing liquid viscosity decreased k(L)a values in the SAR, while k(L)a values in the GIR increased and then decreased with increasing liquid viscosity, following the gas holdup behavior. Increasing liquid density decreased k(L)a in both reactor types. Increasing liquid surface tension increased k(L)a values in the SAR, whereas in the GIR, k(L)a decreased due to the increase of bubble size. Increasing gas diffusivity or gas partial pressure or mixing speed, increased k(L)a in both reactor types. k(L)a values in the GIR were always higher than those in the SAR and increasing D-Imp./D-T and H-F/H-L increased k(L)a in both reactor types. (C) 2003 Elsevier Science B.V. All rights reserved.