We test the rationality of the National Bank of New Zealand's survey data of inflation expectations. We cannot reject the null hypotheses of unbiasedness, efficiency, and orthogonality for a sample from 1985Q1 to 1996Q4. The survey's predictive power is better than those of the random walk and ARIMA models. During the period 1992q1-1996q1, where inflation is low and stable, the predictive power of an ARIMA model is better than that of the survey data, and the predictive power of the survey data is as good as that of the random walk model. These results are not inconsistent with rationality.