Norwegian has two connectives meaning when: da and naar. Da- and naar-clauses have been treated as relative clauses that differ essentially in aspectual features (cf. Faarlund et al. 1997). I show that this view fails to fully account for the data, and I argue instead that da and naar are crucially different in their lexical-syntactic properties: Whereas naar always introduces relative clauses (bound or free), da can only introduce relative clauses that have a lexical head. Da-clauses without a lexical head are subordinate clauses adjoined directly to the matrix clause. Corpus data reveal that the actual aspectual properties of da- and naar-clauses correlate with the status of the clause as relative or non-relative.