Using data on 1,360 savings and loan (S&L) institutions, we compare two non‐parametric methods for measuring efficiency: data envelopment analysis (DEA) and algebraic methods based on Varian (1984). We show that both methods are vulnerable to measurement error, although both theoretically and empirically we find the Varian‐style measures to be less vulnerable. Because we have data on the future insolvency of our S&Ls, we can directly compare the two methods by seeing which does a better job of predicting insolvency (working under the hypothesis that efficiency and insolvency should be negatively correlated). We find that various measures perform very similarly, except for the technical efficiency measure in DEA. Importantly, this last measure frequently yields the implausible result that efficiency and insolvency are positively correlated, possibly because it does not account for factor pri