The pressure difference fluctuation signals of the air-water two-phase flow in a horizontal pipe were analyzed by statistical and fractal theory.Seven parameters

i.e. the averaged value of the pressure difference

standard deviation

skewness，energy ratio

box dimension，correlation dimension and Hurst index were obtained and be used as the characteristic vector of an improved BP neural network with self-adapted learning ratio. Learning form training samples

the network could accomplish the objective identification of the unknown flow patterns.The simulated results showed that the flow pattern characteristic vector which was obtained by statistical and fractal parameters could reflect the difference between various flow patterns and the recognition possibility of the network could reached up to about 93 percent.Moreover

the improved BP neural network had the merits of fast convergence for simulation and avoidance of local minimum.