This study applies nonparametric kernel estimation to the specification of both the drift and volatility functions of a class of stochastic short-term interest rate models. This approach allows us to operate in continuous-time, estimating the continuous-time model, despite the use of discrete data. Three different estimation methods are proposed. We apply these methods to two important financial data. After selecting an appropriate bandwidth for each data set, empirical comparisons indicate that the specification of the drift has a considerable impact upon the pricing of derivatives, through its effect on the diffusion function. Indeed, this impact is more substantial than that reported in the literature.