Prediction models for energy consumption in heat pumps are critical for design, selection, control and fault detection. However, nowadays, the methodology for characterizing variable-speed compressors has not been standardized, and typically the pair compressor-inverter is studied as a whole. In this study, disaggregated energy consumption results of compressor and inverter are provided by using a double-wattmeter scheme. These results are used to compare the performance of existing models and to establish guidelines to define testsmatrixes as compact as possible to characterize this type of compressors. The widely used 20-coefficient correlation can overfit the compressor behaviour when fitted with an small dataset, introducing significant deviations when extrapolating. Optimal Design methodologies to create rating test matrixes gave excellent results compared with classical clustering techniques. Regarding inverter results, a compact empirical model was proposed to model inverter power losses depending on total consumption and compressor speed. The proposed model used only five fitting coefficients, and most predictions are inside a 5% error band for all tested inverters. Finally, the tested compressors and inverters suffered a significant efficiency loss at very low speeds, which could make cycling more profitable under certain circumstances.