The reagent selection is complicated due to high stability of emulsions. It is often ineffective and long. A study was carried out in laboratory settings on several demulsifier samples to identify their effectiveness in dehydration of heavy oil fractions. The comparison of physical-chemical properties of demulsifier reagents and water separation dynamics is presented. A program has been developed to increase the speed of determining the reagent effectiveness in the oilfield. An algorithm of work in the program is presented to introduce machine learning in the oilfield.