Most eukaryotic centromeres are located within heterochromatic regions. Paradoxically, heterochromatin can also antagonize de novo centromere formation, and some centromeres lack it altogether. In order to investigate the importance of heterochromatin at centromeres, we used epigenetic engineering of a synthetic alphoidtetO human artificial chromosome (HAC), to which chimeric proteins can be targeted. By tethering the JMJD2D demethylase (also known as KDM4D), we removed heterochromatin mark H3K9me3 (histone 3 lysine 9 trimethylation) specifically from the HAC centromere. This caused no short-term defects, but long-term tethering reduced HAC centromere protein levels and triggered HAC mis-segregation. However, centromeric CENP-A was maintained at a reduced level. Furthermore, HAC centromere function was compatible with an alternative low-H3K9me3, high-H3K27me3 chromatin signature, as long as residual levels of H3K9me3 remained. When JMJD2D was released from the HAC, H3K9me3 levels recovered over several days back to initial levels along with CENP-A and CENP-C centromere levels, and mitotic segregation fidelity. Our results suggest that a minimal level of heterochromatin is required to stabilize mitotic centromere function but not for maintaining centromere epigenetic memory, and that a homeostatic pathway maintains heterochromatin at centromeres.This article has an associated First Person interview with the first authors of the paper. © 2020. Published by The Company of Biologists Ltd.