The methodology for the repair of critical-sized or non-union bone lesions has unpredictable efficacy due in part to our incomplete knowledge of bone repair and the biocompatibility of bone substitutes. Although human mesenchymal stem cells (hMSCs) differentiate into osteoblasts, which promote bone growth, their ability to repair bone in vivo has been variable. We hypothesized that given the multistage process of osteogenesis, hMSC-mediated repair might be maximal at a specific time point of healing. Using a mouse model of calvarial healing, we demonstrate that the osteo-repair capacity of hMSCs can be substantially augmented by treatment with an inhibitor of peroxisome proliferator-activated receptor γ, but efficacy is confined to the rapid osteogenic phase. Upon entry into the bone-remodeling phase, hMSC retention signals are lost, resulting in truncation of healing. To solve this limitation, we prepared a scaffold consisting of hMSC-derived extracellular matrix (ECM) containing the necessary biomolecules for extended site-specific hMSC retention. When inhibitor-treated hMSCs were coadministered with ECM, they remained at the injury, well into the remodeling phase of healing, which resulted in reproducible and complete repair of critical-sized bone defects in mice in 3 weeks. These data suggest that hMSC-derived ECM and inhibitor-treated hMSCs could be used at optimal times to substantially and reproducibly improve bone repair.