The author compares the performance of three Gaussian approximation methods—by Nowman (1997), Shoji and Ozaki (1998), and Yu and Phillips (2001)—in estimating a model of the nonlinear continuous-time short-term interest rate. She finds that the performance of Nowman's method is similar to that of Shoji and Ozaki's method, whereas the window width used in the Yu and Phillips method has a critical influence on parameter estimates. When a small window width is used, the Yu and Phillips method does not outperform the other two methods. Choosing a suitable window width can reduce estimation bias quite significantly, whereas too large a window width can worsen estimation bias and the fit of the model. An empirical study is implemented using Canadian and U.K. one-month interest rate data.