Model-based engineering (MBE) recognizes models as central in software construction with the possibility of their management in libraries and repositories with proper structuring of their spaces and operations. Due to this success, models (and metamodels) are becoming larger and larger and technics are needed in order to comprehend and exploit them, such as circumscribing sub(meta)models of interest, which is the subject of this paper. Following MBE, there are mainly two ways for circumscribing submodels: only at the model level (by selecting model elements of interest) or through the meta level (by selecting a submetamodeling dimension of interest). In a preceding paper, we deeply studied the first way. Here we concentrate on the second way. Model projection deeply relies on the concepts of submodels and submetamodels with their inclusion qualities for model space structuring and has to be systematically examined from this point of view. It is important to point out that model treatment has to deal with full models (as offered by “off the shelf” libraries) but also with not necessarily well-formed ones, such as unspecified model chunks, due, for example, to the storage in repositories of incomplete engineering choices or of intermediate results of operations. It is a difficulty to encompass all these forms of models, being well-formed or not, in a homogeneous manner through MBE operations. The operation for “Model projection relative to submetamodeling dimensions” presented here does take this difficulty into account.