The mathematical problem of train makeup at rail terminals is formulated. The problem definition is the following. There is a set of cargo ships with goods which must be transported by train. For each cargo unit we know its destination station, cargo ship and importance. For each cargo ship different time intervals which should be spent to reach different seaports are known. It is necessary to makeup train with goods and deliver all these goods with the smallest value of the objective function. Several constraints are set such as all goods from the same ship must be delivered to the same port. In order to obtain the best way for each cargo unit, Dijkstra’s algorithm is used. The problem of locomotive assigning to trains was reformulated as generalized assignment problem with constraints given by three matrices. The main problem of train makeup was solved by the Gurobi optimizer. Numerical experiments were carried out and throughput capacity of the railroad area was estimated.