An arm phantom for testing non-invasive blood pressure measuring instruments under reproducible and real conditions has been developed. Signal records, recorded on patients and validated by a reference method, are played into the instruments to be tested using a combination of an electro-pneumatic and an electro-acoustic converter. For each of 90 patients a database contains records of at least three different kinds of signal: cuff pressure oscillations, Korotkoff sounds (except in neonates) and cuff pressure. As the inflation and deflation rate varies between different sphygmomanometers the control and synchronization of the signal output are ensured as a function of the instantaneous cuff pressure, measured continuously by the system. A segmented output procedure synchronizes the data flow during the test procedure. The records of blood pressure signals are separated in different signal groups, containing only one pressure oscillation and Korotkoff sound respectively, whereas the instantaneous cuff pressure determines the signal group to be selected. Suitable interfaces to the sphygmomanometers under test consist of two specifically developed signal converters, an electro-pneumatic converter and an electro-acoustic converter. The system generates both cuff pressure oscillations and Korotkoff sounds from real data earlier recorded on human subjects. It allows any inflation and deflation rate of the cuff pressure to be applied, considers the variability of the pulse rate occurring in the same data record and finally it allows the superposition of artefacts. The arm phantom can replace clinical trials. The possibility of testing the limits of the performance of sphygmomanometers has been considerably extended.