Background. Speech is the most disturbing noise source in open-plan offices. The disturbance can be reduced by e.g. sound-absorbing materials, masking sound and screens between workstations. An additional reduction in speech level can be achieved by e.g. partially enclosed workstations, pods, sofa groups and chairs, and fully enclosed booths. However, there is no standardized method to declare the sound reduction properties of such ensembles. Aim. Our purpose was to introduce a new method to determine the sound reduction of furniture ensembles. The second aim was to demonstrate which physical parameters affect the sound reduction. Materials and Methods. The test consists of two successive sound power level measurements in a reverberation room. The sound reduction is defined as the level difference without and with the furniture ensemble when a loudspeaker is placed in the occupant's position inside the furniture ensemble. A single-number quantity, speech reduction index, DS, is defined as the difference of the A-weighted sound power levels when the loudspeaker's power is normalized to the standardized speech spectrum. Sixteen furniture ensembles were tested having various parameters with respect to isolation and sound absorption. Results. The values of DS ranged from 0 to 22 dB. Regarding normal workstations, the isolation correlated strongly with DS while the weighted absorption coefficient αw and screen height did not. The measurement uncertainty of DS was 0.9 dB being sufficient to discriminate two ensembles if their DS values differ more than this value. However, typical values for open workstations and partially open workstations seemed to be low. They cannot always be discriminated. Practical implications. A simple method was developed to determine the acoustic performance of furniture ensembles. Test laboratories are encouraged to apply this method to acquire more experience in this regard.