A common assumption about the function of gender agreement in discourse is that it serves referent identification and reference tracking. There is some evidence in favour of this view, and an increasing body of research which finds no foundation for it; this study further supports the latter perspective. It is based on a fully annotated corpus of the Papuan language Mian and uses a noteworthy property of the Mian agreement system: object agreement in transitive verbs is “sporadic”, i.e. it depends on the lexical type of a transitive verb whether it agrees with its object. Therefore we can measure whether speakers of Mian manipulate overt vs. null arguments in discourse to compensate whenever lack of agreement might make argument reference ambiguous. The results clearly show that the proportions of overtly realized objects for agreeing verbs and non-agreeing verbs do not differ significantly, thus lending little support to the claim that gender agreement serves a major function in reference tracking in discourse.