In this paper two different methodologies for model order reduction for thermal problems are analyzed. These methodologies (Arnoldi algorithm and Proper Orthogonal Decomposition) face the problem from two different points of view. The first, deals with systemn matrices obtained from the space discretization of the system of partial differential equations (e.g. finite element method), which describe the relationship between the different nodes of the system, while Proper Orthogonal Decomposition deals with the input-output information of the system. Both methodologies have its own advantages and inconvenients, which will be discussed for a particular MEMS device.