To integrate CAD and CAM of 2½ D prismatic machining parts, part descriptions in low-level format, such as Boundary Representation, are typically interpreted into forms of high-level features, such as pockets and virtual pockets, to facilitate process planning and tool path generation. A recognition procedure is required to transform the low-level data into machinable pocket definition. For complex depression volume where pocket interactions occur, current feature recognition algorithms do not adequately recognize the pockets such that the complex depression volume can be transformed into individual pockets machinable by current pocket milling techniques. Furthermore, in a complex depression, there may exist multiple equally valid set of pockets to describe the same total removable volume. The problem of recognizing multiple interpretations of pockets from part description in Boundary Representation (Brep) is presented in this paper. The algorithm presented in this paper first uses volume decomposition to decompose the total removable volume into small blocks of machinable volume. Then based on machining strategies of 2½D pockets, the small blocks are reconnected in a systematic way to reconstruct pocket volumes and generate multiple interpretations. The result provides multiple sets of pockets, and hence shows that there are multiple ways to machine a part with interacting pockets. Sample results demonstrating the implementation and the capabilities of the algorithms are discussed.