Split DP Topicalization, found in various languages such as German, Romanian, Albanian or Hungarian, looks as a DP with the determiner (usually an indefinite or quantifier) stranded in the base position and its nominal complement fronted to a sentence-initial position, with topic interpretation. This structure is problematic because it seems both to require and to exclude an analysis by movement. Based on evidence from Romanian and German, I show that the movement analysis is correct, and that all the properties which seem to plead against it result from PF processes. The theoretical implications of this analysis are that post-syntactic morphology is necessary, and that certain agreement phenomena, which may be described as "PF-agreement", take place late, inside a phase domain, by the time the phase is completed and the phase domain becomes opaque.