This paper investigates the validity of the Fisher hypothesis in Japanese long-term interest rates (two, three, four, five, seven and ten years) using non-stationary time series models. Initially, the entire sample period (October 1987-June 2006) is investigated. Then the samples, divided into three segments depending on each period's monetary policy regimes, are investigated. Thus, the asymmetric impacts of inflation expectation on Japanese long-term interest rates can be investigated. The first period examined is from October 1987-June 1991. The second period is from July 1991-July 2000. The third period is from March 2001-June 2006. In the first period monetary policy is tightening, but in the second and third periods it is easing. Thus it is concluded that the Fisher hypothesis is valid only in all maturities of long-term interest rates in the first period. However, it does not hold in any of the maturities in the entire sample period or in the second and third periods.