This paper applies the Campbell-Shiller (1987) methodology to a study of the joint behaviour of a three-month and a five-year government yield in the United States, Canada, the United Kingdom, Germany and Japan. The period studied is for most countries the mid-1970s to the third quarter of 1997. The empirical results allow the rejection of the expectations theory of the term structure at high levels of statistical significance in every country except Japan. Furthermore, in every country where the expectations theory fails, the failure of the theory is consistent with the spread overreaction hypothesis of Froot (1989) and Campbell and Shiller (1991). This implies that the departures of long rates from levels predicted by the expectations theory in many major markets cannot be attributed to white noise error terms.