Twin support vector machine~(TSVM) is a powerful learning algorithm by solving a pair of smaller SVM-type problems. However, there are still some specific issues waiting to be solved when it faces with some real applications, \emph{e.g}, low efficiency and noise data. In this paper, we propose a Fast and Robust TSVM~(FR-TSVM) to deal with these issues above. In FR-TSVM, we propose an effective fuzzy membership function to ease the effects of noisy inputs. We apply the fuzzy membership to each input instance and reformulate the TSVMs such that different input instances can make different contributions to the learning of the separating hyperplanes. To further speed up the training procedure, we develop an efficient coordinate descent algorithm with shirking to solve the involved a pair of quadratic programming problems (QPPs) of FR-TSVM. Moreover, theoretical foundations of the proposed model are analyzed in details. The experimental results on several artificial and benchmark datasets indicate that the FR-TSVM not only obtains the fast learning speed but also shows the robust classification performance.