A term structure model bearing features of stochastic volatility and stochastic mean drift with jump (SVJ-SD model for short) is built in the paper to describe the stochastic behavior of interest rates. Based on sample data of an interest rate of national bond repurchase, maximum likelihood (ML), linear Kalman filter and efficient method of moments (EMM) are used to estimate the model. While ML works well for simple models, it may lead to considerable deviation in parameter estimation when dynamic risks of interest rates are considered in them. Linear Kalman filter is a tractable and reasonably accurate technique for estimation cases where ML was not feasible. Moreover, when compared with the first two approaches, using EMM can obtain better parameter estimates for complex models with non-affine structures.