Phase fraction is one of the most important flow parameters for oil-gas-water three-phase flow. However, in traditional research about estimating phase fraction, the complex and unstable flow structure of oil-gas-water three-phase flow makes it difficult to obtain the mapping relationship between phase fraction and sensing data. To estimate phase fraction accurately, electrical and ultrasonic multi-mode sensors are used to detect parameters of the horizontal oil-gas-water three-phase flow including fluid impedance, gas-liquid interface position and water phase distribution information on the section. The eigenvalues of different time series acquired by different sensors are extracted respectively to form a one-dimensional dataset. A fully connected neural network (FCN) based model is established to estimate phase fraction by inputting eigenvectors to the network. Experimental results show that the presented method can realize estimation of phase fraction for oil-gas-water three-phase flow.