Purpose Along with the increase in primary total knee arthroplasty, there has been an increase in the number of revisions. The aim of this study was to propose a selection algorithm for the knee revision constraint according to the state of ligaments and to the bone defects Anderson Orthopaedic Research Institute Classification [AORI] classification. The hypothesis was that this algorithm would facilitate the appropriate choice of prosthesis constraint, thus providing stable components and a good long-term survivorship of the knee revisions. Methods Sixty consecutive revision knee arthroplasties in 57 patients were prospectively evaluated. Prostheses implanted at revision included postero-stabilised, condylar constrained and rotating hinged, relative to the state of the ligaments and of the bone loss around the knee. The median follow-up was nine years (range, 4-12). Results The median IKS knee and function scores and HSS score were 41 (15-62), 21.5 (12-43) and 34 (23-65) points, respectively, before the operation, and 81 (48-97), 79 (56-92) and 83.5 (62-98) points (p<0.001) at the latest followup evaluation. The median ROM increased from 74°(29-110°) preoperatively to 121°(98-132°) (p<0.01) at the final follow-up. Re-revision was necessary in five (8.3 %) patients. Conclusions A selection algorithm for the revision implant constraint based on the state of ligaments and the bone loss AORI classification could provide stable knee reconstructions and long-term success of knee revisions.