In situ tensile tests were carried out during X-ray microtomography imaging of a smooth and a notched specimen of dual phase steel. The void coalescence was first qualitatively observed and quantitative data concerning this damage step was then acquired. The void coalescence criteria of Brown and Embury and of Thomason were then tested against the experimental data at both the macroscopic and local levels. Although macroscopic implementation of the criteria gave acceptable results, the local approach was probably closest to the real nature of void coalescence, because it takes into account local coalescence events observed experimentally before final fracture. The correlation between actual coalescing couples of cavities and local implementation of the two criteria showed that the Thomason criterion is probably the best adapted to predict the local coalescence events in the case of the material studied.