We show that the accuracy of mitotic segregation of three engineered, mapped human mini-chromosomes differs between human, mouse and chicken cell lines. We have studied the cause of these differences by analysing the extent of centromere formation on one mini-chromosome immunocytochemically. In human and chicken cell lines the mini-chromosomes segregate accurately and form centromeres but in one mouse cell line centromere formation is undetectable and mitotic segregation is inaccurate. These results indicate that the centromere is maintained by an activity that functions in trans and varies either in amount or specificity between different cells. Structurally defined mini-chromosomes may allow this activity to be studied.