Knowledge of the onset of fluidization is of considerable relevance and the key to three-phase fluidized-bed reactors design and safe operation. Accordingly, using a wide historic U-Lmf database set up from the open literature, all the quantification methods proposed to predict the minimum fluidization liquid velocity in three-phase fluidized beds have been thoroughly revisited and critically evaluated herein. The database, providing access to diversified information related to over 540 measurements, is dedicated to embracing wide-ranging fluids and bed properties. It covers 30 various particles and 18 liquids and includes data such as aspect ratio, wall effect (or column-to-particle diameter) ratio, and Re-Lmf ranging from 0.8 to.27, 9 to 127, and 10(-2) to 800, respectively. Indeed, the U-Lmf behavior is largely nonlinear and thus cannot be accurately described using the existing empirical and physical approaches. As a result, multilayer perceptron artificial neural networks have been extensively used to generate two highly accurate, a purely dimensional and a dimensionless, empirical correlations describing the U-Lmf Using crosscorrelation analyses, two unsuspected effects, namely, the wall effect ratio and the liquid surface tension, have been unveiled and then incorporated as correlating variables in the neural network correlations. The resulting mean relative error produced by the dimensional correlation is about 16% while the estimated error associated with the dimensionless-based correlation is 30%. The prediction errors from both correlations are found to be insensitive to column-to-particle diameter ratio. Moreover, the neural network approach has been shown to predict with moderate success the minimum fluidization gas velocity, U-Gmf, in liquid-bouyed gas-activated three-phase fluidized beds containing coarse particles (d(v) > 1 mm) at high-input gas fractions.