Abstract The subject-predicate agreement system in Icelandic appears to show sensitivity to the morphological marking of case, instead of the syntactic position of the argument to be agreed with. Furthermore, agreement with the Nominative object appears to be optional and may be disrupted by an intervening Dative argument. This article contributes to the existing discussion by proposing a new empirical generalization about the nature of Long-Distance Agreement (lda), i.e., agreement which occurs in a bi-clausal environment, and its interaction with Dative interventions. Based on the new data from an Icelandic variety called Icelandic B, I argue that lda takes place only if the intervening Dative argument undergoes independently motivated A-movement to the edge of vP. The core idea is that the locus of agreement with the Nominative object is v—its Case licensor: lda arises only if v can probe the Nominative argument in the absence of the Dative argument. The proposed analysis thus accounts for the Icelandic patterns in a strictly derivational and locality-based manner, without any recourse to post-syntactic operations, optionality in agreement or significant modifications in the theory of $$phi $$-feature Agree, thus restoring the Icelandic agreement system to normalcy.