In this study, we propose a method of classifying speech under stress using parameters extracted from a physical model to characterize the behavior of the vocal folds. Although many conventional methods have been proposed, feature parameters are directly extracted from waveforms or spectrums of input speech. Parameters derived from the physical model can characterize stressed speech more precisely because they represent physical characteristics of the vocal folds. Therefore, we propose a method that fits a two-mass model to real speech in order to estimate the physical parameters that represent muscle tension in the vocal folds, vocal fold viscosity loss, and subglottal pressure coming from the lungs. Furthermore, combinations of these physical parameters are proposed as features effective for the classification of speech as either neutral or stressed. Experimental results show that our proposed features achieved better classification performance than conventional methods.