The ultrasonic assisted ethanol ammonium sulfate aqueous two-phase extraction method was used to optimize the extraction process of flavonoids from celery leaves by response surface methodology(RSM) and BP neural network. The effects of four main factors on the extraction rate of total flavonoids were studied. The central composite design experiment was adopted. The data sample output from response surface was used as the input sample of BP neural network， and the characteristics of its interactive algorithm were used to optimize the fitting， so as to achieve the purpose of optimizing the extraction process by combining BP neural network with response surface. The results showed that under the conditions of ultrasonic time of 52.25 min， ultrasonic temperature of 52.5 ℃， solid-liquid ratio of 1 ∶ 44 and ethanol content of 61.47%， the extraction rate of total flavonoids from celery leaves could reach 6.62%. Total flavonoids from celery leaves had scavenging effects on DPPH·， ·OH and ·O₂-， and the maximum scavenging rates were 88.2%， 86.1% and 84.3%， respectively. It could be used as a kind of bacterial inhibitor. The content of luteolin and apigenin in the sample extract was relatively high by LC-MS analysis.