Lipid droplets (LD) are organelles with an important role in normal metabolism and disease. The lipid content of embryos has a major impact on viability and development. LD in Drosophila embryos and cultured cell lines have been shown to move and fuse in a microtubule dependent manner. Due to limitations in current imaging technology, little is known about the behaviour of LD in the mammalian embryo. Harmonic generation microscopy (HGM) allows one to image LD without the use of exogenous labels. Adaptive optics can be used to correct aberrations that would otherwise degrade the quality and information content of images. We have built a harmonic generation microscope with adaptive optics to characterise early mouse embryogenesis. At fertilization, LD are small and uniformly distributed, but in the implanting blastocyst, LD are larger and enriched in the invading giant cells of the trophectoderm. Time-lapse studies reveal that LD move continuously and collide but do not fuse, instead forming aggregates that subsequently behave as single units. Using specific inhibitors, we show that the velocity and dynamic behaviour of LD is dependent not only on microtubules as in other systems, but also on microfilaments. We explore the limits within which HGM can be used to study living embryos without compromising viability and make the counterintuitive finding that 16 J of energy delivered continuously over a period of minutes can be less deleterious than an order of magnitude lower energy delivered dis-continuously over a period of hours. LD in pre-implantation mouse embryos show a previously unappreciated complexity of behaviour that is dependent not only on microtubules, but also microfilaments. Unlike LD in other systems, LD in the mouse embryo do not fuse but form aggregates. This study establishes HGM with adaptive optics as a powerful tool for the study of LD biology and provides insights into the photo-toxic effects of imaging embryos.