A previously reported stereoscopic tracking velocimetry (STV) is further refined and applied to measure directional solidification. The flow involves complicated three-dimensional (3-D) convection, being subject to both buoyancy and surface tension forces in addition to conjugate conduction. For our STV, the 3-D tracking of numerous particles is the most important and challenging process. Here, the performances of the tracking algorithms, which are based on artificial neural networks, are first presented with a brief summary of the STV principles. The 3-D experiment measurements of the convective phenomena are then discussed together with the results from two-dimensional numerical modeling for qualitative comparison.