Identification of Alzheimer's disease (AD) risk loci highlighted the role of microglia in the pathogenies of AD. Nevertheless, whether microglia are beneficial or detrimental in AD is still debated. Several studies tested the contribution of microglia to amyloid pathology by microglia depletion, resulting in seemingly contradictory results. Here, we aimed to unify these conflicting findings by systematically testing microglia depletion at various stages of disease progression in mice.Depletion of microglia by PLX3397 treatment before the onset of amyloid plaques in AppNL-G-F mice resulted in a reduction of both plaque load and numbers. When microglia were depleted after amyloid plaque onset, amyloid levels remained unaffected, but plaques exhibited more diffuse morphology. Importantly, the absence of adaptive immunity in AppNL-G-F/Rag2-/- mice did not alter the amyloid load and microglial response to plaques. We further substantiated the beneficial effect of microglia depletion prior to amyloid deposition using a genetic microglia-depletion model (Csf1rΔFIRE/ΔFIRE; FIRE mice). Lastly, partial rescue of amyloid plaque load was achieved by xenotransplantation of human microglia in FIRE mice.Our observations revealed a bi-phasic response of microglia to amyloid plaques. These findings demonstrate that microglia seed plaques in early disease stages. In later disease stages, microglia compact plaques. These results unify seemingly conflicting findings in the literature and warrant careful consideration of the bi-phasic response while designing drugs targeting microglia.Finally, we introduce an improved human microglia-transplantation model, a powerful tool to investigate microglial risk genes and their effect on amyloid plaque load.