Existing methods to image chromosome segregation errors are not suitable for studying human embryos at advanced preimplantation stages. As chromosomal errors are a leading cause of miscarriage and infertility, it remains unclear whether missegregation arises postfertilization. Here we optimize nuclear DNA labeling via messenger RNA electroporation and apply light-sheet live imaging to reveal chromosome segregation errors immediately before implantation. We show that embryos at advanced preimplantation stages display missegregation, including multipolar spindle formation, lagging chromosomes, misalignment and mitotic slippage. Most lagging chromosomes are passively inherited rather than reincorporated. To trace individual nuclei, we developed an open-source, semi-automated segmentation method using a customized deep learning model optimized for variability in embryo size, shape and signal. With this approach, we find most labeled cells remain externally positioned, consistent with placental rather than inner cell mass fate. Our findings raise questions about clinical uses of preimplantation genetic testing for aneuploidy, while providing broadly applicable imaging and segmentation methods for studying diverse cellular structures in human embryos.