In this study, we have developed, for the first time in the field, a 'dynamic' three-dimensional image reconstruction technique for electrical capacitance tomography (ECT) imaging based on a neural-network multi-criterion optimization (NNMOIRT). This development enables a real time, 3D imaging of a moving object to be realized. The image reconstruction scheme of the 3D ECT is established by introducing a 3D sensitivity matrix into the NN-MOIRT algorithm, developed earlier by the authors. The sensitivity matrices employed are based on 6- and 12-electrode twin-plane cylindrical sensors. The NN-MOIRT algorithm reconstructs simultaneously the image voxels (volume pixels) on 20 x 20 x 20 resolutions from the capacitance data obtained using the twin-plane sensors which surrounds the 3D section of 8 cm in length in the cylindrical columns. The technique is successfully tested over a 3D simulated as well as actual experimental objects. The 3D ECT technique is used to investigate the transient phenomena in the entrance region of a 10 cm diameter column using a single nozzle gas distributor with paraffin liquid (Norpar), air, and 200 mu m glass-beads as flow media. Hydrodynamic characteristics of the gas-liquid and gas-liquid-solid flows, including 3D bubble plume spiral motion, 3D large scale liquid vortex dynamics and real time three-dimensional gas holdup distribution are studied. (c) 2005 Elsevier Ltd. All rights reserved.