We used an experimental model producing an aggressive tissue response associated with implant loosening in humans: a 6 mm polymethylmethacrylate (PMMA) cylinder was pistoning 500 microm concentrically in a 7.5 mm hole, with polyethylene (PE) particles, for 8 weeks. At 8 weeks, the PMMA implant was revised with a titanium alloy (Ti) implant, and an identical primary Ti implant was inserted contralaterally for 4 weeks. With this protocol, we evaluated primary and revision plasma-sprayed Ti implants which were loaded under stable conditions with or without allograft, or under unstable conditions without allograft (bilateral primary and revision implants, n 8 per group, 48 implants in 24 dogs). Revision implants had lower interfacial shear strength, less bone in contact with and adjacent to the implant, and resulted in higher levels of IL-6beta and TNFalpha and lower levels of TGFbeta. In both the revision and primary settings, allograft increased shear strength, stiffness and energy, bone-implant contact, and bone area adjacent to the implant. Unstable implants could not generate a mechanically sound interface, and further exacerbated the difference between primary and revision. We conclude that factors important for improving the fixation of revision implants were bone graft and a stable interface.