OBJECTIVE. To measure the accuracy and determine the positive predictive value (PPV) and negative predictive value (NPV) of data submitted to a statewide surveillance system for identifying surgical site infection (SSI) complicating coronary artery bypass graft (CABG) surgery. DESIGN. Retrospective review of hospital medical records comparing SSI data with surveillance data submitted by infection control consultants (ICCs). SETTING. Victorian Hospital Acquired Infection Surveillance System (VICNISS) Coordinating Centre in Victoria, Australia. PATIENTS. All patients reported to have an SSI following CABG surgery and a random sample of approximately 10% of patients reported not to have an SSI following CABG surgery. RESULTS. The VICNISS ascertainment rate for CABG procedures in Victoria was 95%. One hundred sixty-nine medical records were reviewed, and reviewers agreed with ICCs about 46 (96%) of the patients reported as infected by the ICCs and 31 (91%) of the patients identified with a sternal SSI by the ICCs. In one-third of SSIs, the depth of SSI documented by ICCs was discordant with that documented by the reviewers. Disagreement about patients with donor site SSI was frequent. When the review findings were used as the reference standard, the PPV for ICC-reported SSI was 96% (95% confidence interval [CI], 86%-99%), and the NPV was 97% (95% CI, 92%-99%). For ICC-reported sternal SSI, the PPV was 91% (95% CI, 76%-98%) and the NPV was 98% (95% CI, 94%-100%). CONCLUSIONS. There was broad agreement on the number of infected patients and the number of patients with sternal SSI. However, discordance was frequent with respect to the depth of sternal SSI and the identification of donor site SSI. We recommend modifications to the methodology for National Noscomial Infection Surveillance System-based surveillance for SSI following CABG surgery.