|Abstract As high-speed railway is booming worldwide, the communication system with fast-time varying channel has drawn great attention. The comb pilot based linear minimum mean square error (LMMSE) channel estimator is proved to be an effective method for fast time-varying channel estimation. In this paper, the clustered comb pilot-aided channel estimation for orthogonal frequency-division multiplexing (OFDM) system is discussed, where the time varying channel is approximated by a basis expansion model (BEM). A modified clustered comb pilot structure is proposed and justified to improve the estimation performance compared with the clustered comb pilot proposed by Tang. Based on the complex-exponential BEM (CE-BEM) model, a suboptimal-pilot structure is proposed. In addition, optimal pilot length is analyzed and simulated with a predefined total number of pilots. The simulation results show that the modified clustered comb pilot can greatly reduce the estimation error especially with high Doppler spread. The suboptimal-pilot structure with guard pilot approximation is proven to be competitive. Optimal nonzero pilot lengths for different Doppler spread are obtained by simulation with a predefined channel order and fixed pilot subcarriers.