Multimode interference (MMI) devices operate based on the self-imaging principle. A single mode input from the input waveguide is converted to multiple modes in the interference region. By properly choosing the width and length of this section, energy distribution at the output end forms a specified ratio. Thus the MMI device is commonly used as the optical power divider in the integrated optics. The MMI devices operate at a high bandwidth and are insensitive to polarization and small variation of device dimensions. ¡@ In general, MMI manufactures use commercial simulation software to design MMI devices. However, the beam propagation method (BPM), which is the most commonly used method, assumes small angle approximation and ignores wave reflections. Other more rigorous numerical methods such as finite-difference time-domain (FD-TD) and mode matching methods consume too much computer resources and are therefore can not handle large devices like the MMIs. Therefore, we propose two novel numerical schemes to study MMI devices. The first one called full eigen-mode expansion technique (FEMET) includes all necessary modes but neglects reflection at the dielectric discontinuities. The other method considers all modes traveling in both directions is based on the coupled transverse mode integral equation (CTMIE) formulation. It is most rigorous among all methods and is capable of handling very large device structures. ¡@ In this thesis, we report the demonstration of MMI devices by efficient FEMET and rigorous CTMIE methods. Comparing the two novel studies, the CTMIE approach although highly accurate, it is much more complex and is very difficult to program. It also spends much more computer time than FEMET. On the other hand, FEMET method, being an approximate theory, produces results that are very close to that of the CTMIE results for MMI devices that are correctly designed. We found that either one of the two novel methods are able to compute quickly and accurately for adiabatic MMI devices.