In [2] we presented an algorithm for decomposing a boundary representation model hierarchically into regularity features by recovering broken symmetries. The algorithm adds new recoverable edges and faces, which can be constructed from existing geometry. This generates positive and negative volumes giving simple, more symmetric sub-parts of the model. The resulting regularity feature tree may be utilised for regularity detection to describe a model’s design intent in terms of regularities such as symmetries and congruencies.