The increasing application of systems engineering to the design of modern railways has placed a heavy demand on high quality software-based railway system simulators. Very often, a simulator of this kind is not only expensive to build but also to maintain. Modifications and extensions of an existing simulator are always necessary. This leads to the need for investigation into the use of advanced modelling and software engineering techniques to improve the simulation programs such that they are robust yet easy to change. The work described in this thesis, divided into two parts, investigates a) software design using object-oriented technology; b) algorithms for the efficient solution of power network in the DC railway system simulation context. Regarding the software design, an adaptable simulation framework design based on a subsystem-manager-database structured concept has been built using Borland C++. A class library consisting of 66 classes has also been developed. The simulator developed to date, working on any IBM compatible PCs, is able to produce system performance including substation loads, train voltage and current profiles, rail potentials and train diagrams. The simulator models have been verified by means of comparing the results generated by the Birmingham University Fortran multi-train simulator. On the efficient power network solution algorithms, an extensive investigation into the sparse matrix and iterative numerical methods has been conducted. Several representative algorithms have been coded for a comprehensive dynamic speed trial. According to the results, the variable bandwidth preordering is by far the most efficient algorithm for small to medium scale simulations, whilst the minimum degree ordering is the fastest algorithm for simulations including rail potential calculations in which the system usually has several hundred nodes.