By integrating a varying coefficient model with a GARCH (1,1) model, this study develops a factor-dependent interest rate model framework which is able to dynamically adjust the parameters of the model to reflect both the changes of the factor of the macro-economy and the effects of volatility clustering. Using American one-month treasury rate, empirical results of this study show that the proposed factor-dependent models outperform fixed-parameter models in different shapes of term structure (downward-sloping, upward-sloping, and flat) and in different estimation periods. Additionally, this study also finds that the term-spread variable is more informative when term structure is downward-sloping.