Risk models are useful in evaluating and comparing surgical outcomes, but surgeons may not always agree with the risk estimates derived from these models, particularly in high-risk cases. We examined the concordance between surgeons' and a risk model's predictions of operative mortality in high-risk coronary artery bypass grafting (CABG) patients, and we attempted to identify the reasons for any discrepancies. From the Veterans Affairs Continuous Improvement in Cardiac Surgery Program (CICSP), a prospective database and cardiac surgery risk model, we obtained data regarding 181 high-risk, isolated CABG cases performed at a single institution between April 1998 and April 2008. Cases were considered high risk if the surgeon estimated the patient's operative mortality risk to be ≥ 10%. We compared the mortality predictions made by surgeons and the risk model by using the signed-rank test and investigated cases in which there was a significant discrepancy (at least 2-fold) between the two predictions. The observed 30-d/in-hospital and 180-d mortality rates were 6.1% (11/181) and 11.0% (20/181), respectively. The mean operative mortality prediction made by surgeons (12.0% ± 5.3%) was higher than that made by the risk model (7.5% ± 8.5%) (P < 0.001). There was significant discrepancy between the surgeon and risk model estimates in 62% (113/181) of cases. In 53% (60/113) of these cases, the surgeon reported having considered risk factors not included in the CICSP model, including (most commonly) possible need for an additional procedure (n = 15), compromised mobility (n = 11), liver disease (n = 9), hematologic or immunologic disease (n = 6), and quality of targets (n = 5). In high-risk CABG cases, surgeon and CICSP risk estimates often disagreed markedly, partly because some disease entities of concern to surgeons are not included in the risk model. The higher mortality risk estimated by the surgeons is a better reflection of the considerable mortality risk that extends up to 180 days after surgery.