The Abductive Network (AN), a group method data handling (GMDH) algorithm-based self-organizing model, was applied to study the solids holdup distribution of a liquid-solid circulating fluidized bed (LSCFB) system. At first, the AN model was trained with experimental data sets obtained from a pilot scale LSCFB operated with spherical glass beads and irregularly shaped lava rock particles as solid phases, and water as liquid phase. In the model training the effects of various auxiliary and primary liquid velocities and superficial solids velocities on radial phase distribution at different axial positions were considered. The developed AN model was employed to predict the solids holdups at various locations of the LSCFB riser under different operating conditions. The competency of the developed AN model was confirmed by comparing the model-predicted and experimental solids holdups of the LSCFB system. Both the experimental and model-predicted outputs showed that under different superficial liquid velocities the solids holdup was higher for the glass beads than the lava rocks. The solids holdup decreased with increasing liquid velocity at all axial locations. The radial non-uniformity distributions of solids holdup in the central region decreased toward the wall. The higher drag force acting on the spherical-shaped particles was likely responsible for this variation of the solids holdups.