Kinetochores can form and be maintained on DNA sequences that are normally non-centromeric. The existence of these so-called neo-centromeres has posed the problem as to the nature of the epigenetic mechanisms that maintain the centromere. Here we highlight results that indicate that the amount of CENP-A at human centromeres is tightly regulated. It is also known that kinetochore assembly requires sister chromatid cohesion at mitosis. We therefore suggest that separation or stretching between the sister chromatids at metaphase reciprocally determines the amount of centromere assembly in the subsequent interphase. This reciprocal relationship forms the basis of a negative feedback loop that could precisely control the amount of CENP-A and faithfully maintain the presence of a kinetochore over many cell divisions. We describe how the feedback loop would work, propose how it could be tested experimentally and suggest possible components of its mechanism.