In this paper I shall address the form and distribution of subject agreement markers in Udi, a Caucasian language spoken in the Republics of Azerbaijan and Georgia (former USSR). Subject-verb agreement in Udi, just like in most other languages, is a property strongly associated with the verb (or predicate): in Udi, the lexical class of the verb determines to a large extent the allomorphy observed in the agreement paradigm. Furthermore, the verb, and proper parts of it, can also serve as a host, to which the agreement marker can attach, sometimes exhibiting infixation. In the presence of negation, wh-expressions, and focussed constituents, however, realisation of subject agreement on the verb is illicit. Instead, the agreement marker attaches to one of the aforementioned syntactically independent elements. Even when separated from the verb, these markers exhibit more or less the same allomorphy as if they were attached to the verb. I will argue that this morphosyntactic paradox can be resolved, if we assume that word-level sings can contribute more than one domain object on their DOM list (Kathol, 1995). Udi agreement markers are, then, introduced in the lexical entry of the verb, accounting for their morphological properties, while at the same time, they will enjoy a representation as an (independent) domain object, enabling us to capture their placement properties. In particular, I will build on the proposals for a linearisation-based analysis of morphosyntactic paradoxa in Fox and European Portuguese (Crysmann, 1999b; to appear) and show how the ideas developed there can be fruitfully employed in the analysis of Udi as well. It is of note that the class of elements to which Udi agreement markers attach is actually quite similar to the class of syntactic elements triggering proclisis in European Portuguese.