In Taiwan, railroad has an extremely high demand during holiday peak periods. The traffic pattern in holidays, such as time of demand and average travel distance, are very different from that of the normal days. In order to develop a better train schedule for holiday travelers, this paper first determined the train passengers' preference pattern from the historical booking data of Taiwan Railway Administration (TRA). Based on the concept of market separation, the railroad's passenger market during long-weekend holidays was divided into three different categories according to the distance of travel. Passengers in each category are assumed to have different demand characteristics. Different traveling time, differnet departure time, and the availability of seat are used as the three basic factors to develop the decision function for travellers in each category. Questionnaire surveys were conducted on passengers in selected train stations during two holiday periods in 1997. The general choice model (GCM) was used to analyze the responses. Twenty-seven models for three different classes of train service and three different market sectors were calibrated. Sensitive analyses were also performed on the preference models developed in this study. The results indicated that passengers in different market segments have significant different preferences with respect to service characteristics. For example, medium- and long- distance passengers are more sensitive to the change of travelling time, the availability of seat, and the class of service than short distance passengers. It is hoped that the passenger preference models can be used as a useful tool for TRA to plan and improve the train scheduling in holidays and to better utilize its facilities as well.