||In Hong Kong, ferry services play a supplementary role in serving the inner harbor traffic and provide essential links to the outlying islands where land transport alternatives are not available. In recent years, due to the increasingly comprehensive and efficient road and rail networks, and the expansion of land-based public transport, ferry patronage has declined steadily. The government, on the one hand, plays an important role to facilitate the creation of a financially feasible operation environment (operator objective). On the other hand, the government is responsible for maintaining an affordable and quality ferry service to the public in terms of speed, comfort, and frequency (user and government objectives). The current practice of the Hong Kong government is to carve ferry services into several packages: combing those routes that are profit making and those that are losing money, then each service package is operated by one company. This practice enables cross-route subsidies so as to keep socially desirable albeit money-losing routes. However, the current status is that service providers operate the route sets independently, in terms of service schedule, ferry types, and fleet size. It is, therefore, beneficial to develop a more advanced and systematic fleet management to achieve route interlining and timetable synchronization so as to reduce operation and capital costs. In this study, a ferry fleet management model is developed for determining what are the optimal mixed ferry fleet size, routes and schedules on multiple routes that will minimize the total system cost. Five numerical examples are used to demonstrate different ferry operation scenarios which are categorized by ferry type (ordinary and fast), trip nature (non-stop and multi-stop) and operation environment (government objective consideration). A two-phase heuristic solution algorithm is developed to solve a much larger-scale problem. In this algorithm, Phase I is to determine a feasible upper bound of the optimal solution, whereas Phase II is to search for an upper bound improvement. This heuristic is applied to solve two scenarios and the result shows that the approximation error is below 4%. The time of heuristic implementation for the above two scenarios is less than 40 seconds, whereas the running time consumed by CPLEX is 8835 seconds in Scenario I and 6525 seconds in Scenario II. It demonstrates that the heuristic can solve the problem efficiently.