It has been argued that controlling voices in different vocal registrations (passaggio; Italian) is one of the important techniques in the operatic singing. However, critical acoustic features in evaluation of skillfulness of the passaggio has remained unclear. In this study spectral variations between voices in scale sequences were calculated based on recordings of 20 graduate level students majoring in vocal music. Perception evaluation was carried out for these recordings with 15 graduate students in vocal music as evaluators. A general linear regression analysis-the average RMS errors between the smoothed spectra of consecutive tones was a significant factor to predict the perceptual evaluation. It was also found that the relative duration of the initial note in each scale singing against the remaining notes could affect the perceptual evaluation. This factor significantly interacted with a spectral deviation from the average spectrum of each singer, i.e., a group of singers with shorter durations tended to be evaluated better when the degree of spectral deviation was small.