This paper describes a new canine segmental spinal fusion model for the comparison of bone-grafting materials. The test sites in the model are three separate posterior interfacet-interlaminar fusion sites in the lumbar spine (L1-2, L3-4, and L5-6). The outcome was assessed by scoring of the fusion sites for union and by mechanical testing of fused segments. The results from two experiments are presented. Autogenous cancellous bone was the most effective material tested and had a statistically superior score for union compared with all other materials (p = 0.01). The results with an osteoconductive matrix of collagen and ceramic alone were no better than those with the controls (no graft). However, addition of an extract of matrix-derived proteins (15-30 kDa) to the collagen-ceramic carrier appeared to improve the score for union. The inclusion of nonresorbed ceramic granules had no evident effect on the mechanical properties of fusions, with a comparable score for union. This model appears to be a sensitive and efficient method for the comparison of graft materials. Advantages over previously described models are discussed.