A study was performed in dogs to evaluate the dose-response characteristics and effectiveness of recombinant human bone morphogenetic protein-2 with a collagen sponge carrier in a segmental defect model. Twenty-seven dogs underwent bilateral radial osteotomies with creation of a 2.5-cm diaphyseal defect. All received autogenous cancellous bone graft in one defect and a collagen implant in the other. These implants contained recombinant human bone morphogenetic protein-2 at the following doses: group 1 at 0 microg (three dogs, 0 microg/ml total implant volume), group 2 at 150 microg (three dogs, 50 microg/ml), group 3 at 600 ,g (three dogs, 200 microg/ml), group 4 at 2,400 microg (three dogs, 800 microg/ml), group 5 at 0 microg (five dogs, 0 microg/ml), group 6 at 150 microg (five dogs, 200 microg/ml), and group 7 at 600 microg (five dogs, 50 microg/ml). The defects were stabilized with external fixators. The dogs in groups 1-4 were killed at 12 weeks postoperatively, and those in groups 5-7 were killed at 24 weeks postoperatively except for one dog in group 7, which was killed at 48 weeks. Evaluation included monthly radiographs, biomechanical testing, and nondemineralized histology. All 27 radii with autogenous cancellous bone graft and all 19 implants treated with recombinant human bone morphogenetic protein-2 achieved radiographic and histologic union and gross stability. The eight radii treated with collagen carrier alone went on to radiographic and histologic nonunion and were grossly unstable at death. A dose-dependent occurrence of cyst-like bone voids was noted radiographically and histologically. Biomechanical performance tended to be better at the lowest dose studied at 12 weeks, and all three doses performed better than the placebo (p < 0.05) at 12 and 24 weeks. By 24 weeks, radiolucent areas corresponding to histologic bone voids persisted radiographically, although there was evidence of early bone remodeling. This remodeling progressed to 48 weeks in the single animal followed to this time point, although bone voids remained. These radiologic findings were confirmed histologically. Recombinant human bone morphogenetic protein-2 in a collagen sponge carrier has significant osteoinductive activity in this canine segmental defect model. A dose-response relationship is evident, with heterotopic bone and cyst-like void formation at higher doses and a minimum effective dose of 0-150 microg. At 12 and 24 weeks postoperatively, biomechanical parameters achieved by defects treated with recombinant human bone morphogenetic protein-2 were comparable with those of autograft controls and were significantly stronger than those of the placebo (p < 0.05).