The paper presents the design and evaluation of an adaptive signal processing procedure based on human skill. The focus is on interpreting probe signals detected in gas–liquid flow in the presence of noise where existing signal interpretation techniques may encounter difficulties. Interpretation of a probe signal requires construction of a corresponding two-state signal that denotes the presence of the phases, i.e. gas and liquid, at the probe tip. To develop a computer procedure that would imitate a skilled operator in probe signal interpretation, manual knowledge acquisition and evolutionary optimization were employed. First, a prototype signal interpretation procedure based on operator skill was designed, and the procedure parameters were then optimized with a genetic algorithm. In the optimization process, a two-state signal reconstructed from the probe signal by an operator was used as a reference. The robustness of the approach was tested in a series of numerical experiments. They included local evaluation on training and test signals, calculation of global void fraction values, and an assessment of variability among different experts. The results showed that the developed technique is highly consistent with the operator way of signal interpretation and represents a reliable prerequisite for gas–liquid flow measurements.