In contrast to the endoscopic examination of a stationary phonation, the examination of a nonstationary phonation screens a broader spectrum of vocal fold oscillations. For this reason, vocal fold vibrations are investigated and recorded with a highspeed camera system in eight normal and in eight pathological voices during a pitch raise. A quantitative analysis of the observed vocal fold dynamics in terms of symmetry and regularity is done with a time-dependent two-mass model of the vocal folds. The model's parameters are numerically optimized to emulate the observed non-stationary vocal fold vibrations. These parameters permit an objective interpretation of vocal fold oscillating asymmetries and allow a classification in normal and pathological cases. The practicability of the optimization algorithm is demonstrated with a set of 242 synthetically generated data sets. By applying the optimization procedure to the recordings of the 16 subjects a correct classification into groups of normal and pathological cases was achieved.