The main objective of this dissertation is to model the Brazilian interest yield curve using Diebold and Li (2006) framework, in order to produce short, medium and long-term forecasts. We estimate the model and then compare its term-structure forecasts with forecasts based on standard benchmark models. Our results lead to the conclusion that the model proposed by Diebold and Li is not consistent with The Brazilian specific evidence, since it is outperformed by simple univariate models, for all forecast horizons, with any maturity choice. We make some theoretical conjectures to explain why the attempt has failed, which inspired the development of two new variants of the original model. The new results indicate that the model improvements proposed are promising, because one of the variants succeeds in producing long-term forecasts of greater accuracy than those based on competing models.