The research presented here uses theoretical constructs of formal syntax to account for performance data in agreement production. The phenomenon examined is object interference in French, i.e., incorrect agreement of the verb with the object. In the first experiment, interference is shown to occur in object relative clauses despite the absence of any surface intervention on the subject–verb relation (e.g., *John parle aux patientes que le medicament guérissent; *John speaks to the patients-P that the medicine-S cure-P). Critically, no interference is found in complement clauses, superficially identical to the relatives (e.g., John dit aux patientes que le medicament guérit; John tells the patients-P that the medicine-S cures-S). Experiments 1–4 tease apart the role, in interference, of three properties that distinguish relative clauses from complement clauses: argumenthood, participle agreement and movement. Results suggest that interference is caused by object movement, and more particularly by the intervention, on agreement, of the intermediate trace of the moved object postulated in theoretical syntax to account for independent phenomena.