This paper deals with a new method of metamodel construction for electromagnetic compatibility (EMC) simulation of power electronic converters. The proposed method uses a compressed wavelet coefficients vector of a time-domain signal and a learning method [support vector regression (SVR)]. This can be applied for a black box approach of modeling of complex systems such as power converters. This allows the study of EMC behavior, the impact of the uncertainties and their propagations inside the model or the design of EMC filters.