In this paper we study the term structure of interest rates and test the rational expectations hypothesis using single regression equations and then multivariate regression equations. Single regression equations are found to produce results that are sensitive to outliers due to finite sample. Multivariate regression equations produce results that are less sensitive to outliers due to a larger sample size, and in our sample, yield a borderline rejection of the rational expectatons hypothesis. We apply a distance covariance test statistic measuring the deviation from independence between the forward forecast errors and present information variables. This measure is asymptotically distributed to be bounded below by χ21 for usual ranges of critical region, and does not require any distributional assumption. The rational expectation hypothesis is more clearly rejected using the distance covariance metric. There is thus preliminary evidence that distributional and linearity mis-specification of the rationality hypothesis in the term structure could potentially biased toward non-rejection of an otherwise generally unsustainable hypothesis.