We present a model for complex documents possibly consisting of a hierarchically structured set of images or texts. Documents are represented both at the form level (as sets of physical features of the representing objects), at the content level (as sets of properties of the repreaentedentities), and at the structure level. A uniform and powerful query language allows queries to be issued that transparently combine features pertaining to form, content and structure alike. Queries are expressions of a (fuzzy) logical language. While that part of the query that pertains to (medium-independent) content is "directly" processed by an inferential engine, that part that pertains to (medium-dependent) form is entrusted to specialised document processing procedures linked to the logical language by a procedural attachment mechanism. The model thus combines the power of state-of-the-art document processing techniques with the advantages of a clean, logically defined framework for understanding multimedia document retrieval.