We propose a semiparametric method for estimating ROC surfaces for continuous diagnostic tests based on two test measurements. Such a three-class diagnostic problem based on two test measurements arises naturally from some DNA amplification-related diagnostic scenarios. Simulation results show that our proposed semiparametric ROC surface estimator is more efficient than the nonparametric counterpart and is quite comparable with the parametric counterpart when model assumption of the data is correctly specified for the parametric method. Moreover, when the parametric model assumption is not true, our proposed semiparametric method is superior to both nonparametric and parametric methods. Some discussions and a simulated example along with R programs for implementation of the proposed method are also provided.