Introduction. Forecasting the term structure of interest rates is considered to be the complex problem. The experience shows that just a few models from the proposed in science literature provide us with much better accuracy in forecasting than random walk. This paper is devoted to the comparison of forecasting results’ accuracy for the different specifications of econometric models using Russian data for 2004–2014 years. Models. We use the next econometric models for the forecasting: random walk, autoregressive model, factor model, Diebold-Li model. Results. Comparison on the Russian government bond market data for 2004–2014 years shows us that autoregressive models are better for the short-term rates’ forecasting, but the dynamic Diebold-Li model gives better accuracy for the mid-term and long-term rates’ forecasting.