This paper presents a feature extraction technique based on the Hilbert-Huang Transform (HHT) method for emotion recognition from physiological signals. Four kinds of physiological signals were used for analysis: electrocardiogram (ECG), electromyogram (EMG), skin conductivity (SC) and respiration changes (RSP). Each signal is decomposed into a finite set of AM-FM mono components (fission process) by the Empirical Mode Decomposition (EMD) which is the key part of the HHT. The information components of interest are then combined to create feature vectors (fusion process) for the next classification stage. In addition, classification is performed by using Support Vector Machines (SVM). The classification scores show that HHT based methods outperform traditional statistical techniques and provide a promising framework for both analysis and recognition of physiological signals in emotion recognition.