Water model experiments were performed in a full-scale, delta-shaped water model tundish, in order to study the removal of inclusions by micro-bubbles. Micro-bubbles were generated using a specially designed ladle shroud with twelve laser-drilled orifices. Gas flow rates, injection positions and multi-port injection were all taken into consideration to create different bubble conditions. Bubbles were recorded using a high speed camera and post-processed with commercial software, Image J. Hollow glass borosilicate microspheres, smaller than 100 μm, were used to simulate inclusions, and detected, in-situ, using a new generation of the Aqueous Particle Sensor, APS III. The results revealed that the effect of microbubbles on inclusion removal depends greatly on the gas injection protocols used. The optimum gas flow rate was an intermediate value, which indicates a minimum particle number density, n p , of about 7.85/ml. This results from the counter-balancing effects of bubble sizes against the total number of bubbles. The highest inclusion removal rate was 80%, when gas was injected through the four ports located closest to the slide gate, at a gas flow rate of 0.2 L/min.