The purpose of this paper is to develop a dynamic thermal model of a permanent magnet excited synchronous motor (PMSM). The model estimates the temperature at specific points of the machine during operation. The model is implemented using thermal network theory, whose parameters are determined by means of analytical approaches. Usually thermal models are initialized and referenced to room temperature. However, this can lead to incorrect results, if the simulations are performed when the electrical machine operates under “warm” conditions. An approach is developed and discussed in this paper, which captures the model in critical states of the machine. The model gives feedback by online measured quantities to estimate the initial temperature. The paper provides an extended dynamic thermal model, which leads to a more accurate and more efficient thermal estimation.