Since the temperature response of an electric machine under a constant duty can be obtained by a single experiment, it is the variable driving cycles that constitute the exact challenge in real-time thermal modeling. This article proposes an excitation-based methodology for real-time temperature estimation. Great detail is presented for the implementation procedure specific to estimating the end-winding temperature of a water-cooled interior permanent-magnet synchronous motor (IPMSM). In this methodology, a linear time-invariant (LTI) representation is devised for the speed-dependent parameter-varying heat-transfer problem of electric machines, followed by a generic solution via Duhamel’s theorem. Each of the response functions in the Duhamel’s integral thermal model is represented by a well-defined auxiliary problem. Computational fluid dynamics (CFD) simulations with calibrated parameters can be conducted to solve these auxiliary problems for the response functions. Because the CFD model features the least equivalent parameters that need experimental calibration, the present methodology can make the most of the geometry and material information and greatly reduce the experimental efforts compared to a data-driven method. Moreover, the present methodology can avoid the need for solving ill-posed inverse heat-transfer problems, and thus the resulted real-time thermal model is pretty reliable and robust.