Predicting the transition from cavity growth to coalescence is important for understanding and simulating ductile failure. Many models have been published and are generally verified by finite element calculations on structures consisting of a periodic network of single defect cells (e. g. [1]). While this approach is appropriate for low porosity, it may underestimate the interaction between defects at higher porosity, in particular by constraining the orientation of the location bands. This study therefore aims to study the transition between growth and coalescence by considering the plasticity behaviour of a cell with a random population of defects. Similar to the numerical homogenization in Fritzen’s boundary analysis ([2]), a population of spherical defects of the same radius is randomly generated within an elasto-plastic matrix (with and without strain hardening). Its behaviour is then studied by finite elements in large deformations for conditions with periodic limits, at different levels of imposed triaxiality. This study is repeated on different defect population draws, in order to be able to statistically analyze the behavior until coalescence. A criterion for the beginning of coalescence may then be proposed.