This paper describes a reinforcement learning (RL) approach to train timetabling, which takes into account the characteristics of inter-city high speed railway lines in China. A potential advantage of the proposed approach over well-established mathematical programming approaches lies in that it does not rely heavily on domain expertise to define the various timetabling rules and strategies. Specifically, a discrete time Markov Decision Process (MDP) is established to model the studied problem, and a well-designed RL method is proposed to solve the problem, assuming that the fundamental information about the studied lines (minimum running times, headways, stopping patterns, etc.) is known. Four inter-city high speed railway lines that operate on the Beijing-Tianjin corridor are employed as a case study to test the performance of the proposed approach. The obtained results preliminarily demonstrate the effectiveness and applicability of the proposed approach.