The paper deals with the difficulties of reconstructing a hydrodynamic structural function from a time‐dependent raw signal produced by a sensor. In the presence of two fluid phases, the deformation of a phase interface at a sensor leads to a lag between the occurrence and the detection of the event, and the probe response characteristics may be altered with time for several reasons. A new phase discrimination technique is presented based on man–machine interaction and utilizing genetic optimization. To discriminate the phases, two‐point threshold discrimination is proposed at first. In addition, a man–machine method is introduced which can mimic a human expert observation of a raw signal and his intuitive discrimination procedure. Finally, genetic algorithms are used to search for those threshold settings that produce solutions closest to those defined by an expert. The proposed new method is capable of adaptation to specific flow conditions. ©1995 American Institute of Physics.