Abstract: Based on dynamic measurement characteristics of turbine flowmeter and radioactive water holdup-densitometer combination tool with basket concentrating flow diverter in oil-gas-water three-phase flow loop，a soft measurement model of artificial neural network (ANN)was established to predict the total flow rate and component flow rate fraction of three-phase flow.The effect of oil-water flow pattern characteristics on the soft measurement model was accounted for by considering the nonlinear leakage of basket concentrating flow diverter and various complex flow pattern variations of the measuring channel after the concentrating flow.Levenbery-Marquardt nonlinear damp least square algorithm was used to train and learn in the model.The model verification results showed that the model could be used to give good prediction accuracy of total flow rate and component flow rate fraction for the bubble and slug flow patterns.It provides an effective information processing method for three-phase flow logging.