The precision of appraisals used to refinance mortgages is particularly difficult to evaluate because there are no sales prices to serve as a basis for comparison. This paper estimates the accuracy of refinance appraisals using a novel methodology that does not require using a sales price as a benchmark: specifically, comparing the performance of appraisals against automated valuation model forecasts to predict loan default. By using a modification of the Vuong (1989) test for non-nested model performance, ranges of the forecast standard errors can be identified in which (1) model valuations outperform appraisals; (2) the performance of both approaches is similar; and (3) appraisals are superior. A second contribution of this paper is to document the improvement in appraisal precision after the Home Valuation Code of Conduct (HVCC) became effective. Our preferred estimates indicate that in the context of refinance transactions over the January 2000–April 2009 period before the HVCC was adopted, appraisals are as precise as a model forecast, with a standard error of approximately 19 percent. In the post-HVCC period, this precision improves to about 12 percent.