We carried out a prospective study to determine whether the addition of a recombinant human bone morphogenetic protein (rhBMP-2) to a machined allograft spacer would improve the rate of intervertebral body fusion in the spine. We studied 77 patients who were to undergo an interbody fusion with allograft and instrumentation. The first 36 patients received allograft with adjuvant rhBMP-2 (allograft/rhBMP-2 group), and the next 41, allograft and demineralised bone matrix (allograft/demineralised bone matrix group). Each patient was assessed clinically and radiologically both pre-operatively and at each follow-up visit using standard methods. Follow-up continued for two years. Every patient in the allograft/rhBMP-2 group had fused by six months. However, early graft lucency and significant (> 10%) subsidence were seen radiologically in 27 of 55 levels in this group. The mean graft height subsidence was 27% (13% to 42%) for anterior lumbar interbody fusion, 24% (13% to 40%) for transforaminal lumbar interbody fusion, and 53% (40% to 58%) for anterior cervical discectomy and fusion. Those who had undergone fusion using allograft and demineralised bone matrix lost only a mean of 4.6% (0% to 15%) of their graft height. Although a high rate of fusion (100%) was achieved with rhBMP-2, significant subsidence occurred in more than half of the levels (23 of 37) in the lumbar spine and 33% (6 of 18) in the cervical spine. A 98% fusion rate (62 of 63 levels) was achieved without rhBMP-2 and without the associated graft subsidence. Consequently, we no longer use rhBMP-2 with allograft in our practice if the allograft has to provide significant structural support.