In order to identify the flow pattern automatically and accurately, a new identification method was developed. Flow images were captured by a high speed CCD in a horizontal pipe. The characteristics of bubbles such as area, width and height were obtained using image processing techniques. First, a fuzzy reasoning method was used to identify stratified flow and annular flow. Then a fuzzy neural network was used to identify the flow pattern of bubbly, slug and plug. Levenberg-Marquart optimized algorithm was used to learn the network, and its constringency is rapid. The experimental results show that the fuzzy method can accurately identify the flow patterns in a horizontal pipe.