Introduction. Wound surveillance has been reported to result in a significant fall in the incidence of wound sepsis in total knee arthroplasty (TKA). However, there is currently little guidance on the definition of surgical wound infection that is best to be used for surveillance. The purpose of this study was to assess the agreement between three common definitions of surgical wound infection as a performance indicator in TKA; (a) the CDC 1992 definition, (b) the NINSS modification of the CDC definition and (c) the ASEPSIS scoring method applied to the same series of surgical wounds. Methods. A prospective study of 500 surgical wounds in patients who underwent knee arthroplasties between May 2002 and December 2004 from a single tertiary centre were assessed according to the different definitions of surgical wound infection. Results. A total of 500 wounds were assessed in 482 patients. Mean age of patients was 70+/−11 years, 61.6% were females, duration of surgery was 101+/−49 minutes and mean follow-up was 35.2+/−25.7 months. The most commonly isolated species were Coagulase negative staphylococci (33.3%), Staphylococcus aureus (25%) and Pseudomonas aeruginosa (16.6%). The mean percentage of wounds classified as infected differed substantially with different definitions: 5.8% with the CDC definition, 3.6% with the NINSS version and 2.2% with an ASEPSIS score > 20. When superficial infections (according to CDC category) were included, 5.2% (26) of all observed wounds received conflicting diagnoses, and 1.4% (7) were classified as infected by both ASEPSIS and CDC definitions. When superficial infections were excluded, the two definitions estimated about the same overall percentage infection (2.2% and 2.6% respectively), but there were almost three times as many conflicting infection diagnoses (n=14) as concordant ones (n=5). Conclusion. Distinctions in surgical wound infection definitions contribute to notable differences in how infections are classified after TKA. Even small changes made to the CDC definition, as with the NINSS version, caused major variation in estimated percentage of wound infection. A single definition used consistently can show changes in wound infection rates over time at a single centre. However, differences in interpretation prevent comparison between different centres