The main goal of this work is to provide engineers with a tool to model the thermal behaviour of electric machines with high accuracy and at the same time low computational effort by means of lumped parameter thermal networks (LPTN). Such a thermal model is also predestined for an application in iterative schemes like parameter studies or design optimizations. The fundamentals of heat transfer and the basic concept of LPTN modeling are summarized. A quadrangular prism shape is introduced. A tableau method based solver using the state-space representation of the thermal system is proposed, which can also be utilized for the identification of unknown thermal model parameters. Two approaches to enhance thermal analysis with lumped parameter networks will be proposed and assessed in detail. The first enhancement is a segmentation-based approach to increase the model resolution in general. By increasing the number of nodes representing a single component, the steady-state as well as transient accuracy can be improved significantly. In addition to that, the distributed nature of the multi-node setup generally allows for the consideration of distributed losses as well. The application to PMSM machines with different stator topologies is demonstrated. The second enhancement is a triangulation based method allowing to model slots with arbitrarily distributed conductors with the LPTN approach. Instead of base bodies, the slot interior is represented by a web of triangle shapes. The thermal resistances between the nodes within these triangles are identified by means of static FEA simulations. Thus, the method allows the modeling of arbitrary conductor arrangements and at the same time a high precision of the results can be ensured.