According to questionnaire studies, speech is the most annoying sound source in open-plan offices. The most distracting speech comes from the nearest workstations. Therefore, speech privacy between workstations should be as high as possible. This presupposes that the speech is properly attenuated and masked. The aim of this study is to present a simple and validated model that can predict the speech transmission index, STI, between two nearby workstations. The speech level propagating different paths and masking noise level from different sources are calculated using conventional room acoustical principles. Speech through the screen, reflected speech via ceiling, diffracted speech over the screen and reverberant speech were involved. Masking noise sources comprise, e.g., ventilation, office equipments, and artificial masking system. Speech intelligibility was determined from the speech-to-noise ratio using the Modulation Transfer Function method, which gives the speech transmission index and the rapid speech transmission index, RASTI. The model was validated using 30 randomly selected workstations in situ. The prediction accuracy was −0.03±0.04. Because the standard error of RASTI measurements was approximately ±0.02 in constant background level, the prediction accuracy of the model was found satisfactory.