In this paper we sift the evidence for overdifferentiated gender marking on numerals in Medieval Italo-Romance. While the dialects of most of Northern Italy still preserve a masculine vs feminine contrast on 'two' and 'three' to this day, their ancestors in the Middle Ages featured an additional third form of each ("doa", "trea"). These forms, it is argued, are not innovations - as some have claimed - but are rather the legitimate outcome of previously regular neuter agreeing forms which can be traced directly back to (late) Latin. By the time the earliest extant documentation of Northern Italo-Romance emerged, these neuter forms of numerals were quite isolated in their respective systems, as neuter agreement had long been eliminated on other agreement targets. Such isolation made the occurrence of "doa" (alongside masc. "doi" and fem. "doe") a case of overdifferentiated gender marking.