Abstract: Back-propagation based improved training network algorithm was proposed and a prediction model for converter end point phosphorus based on fuzzy neural network has been established based on melting process and production data of a 150t converter and analysis on influence factors on end point phosphorus, in accordance with the deficiencies of present back propagation algorithm. The results showed that percentage of hits of converter end point phosphorus content in steel with error ±0. 002% was 68. 69%, that with error 士 0. 004% was up to 95. 96%, and the maximum error of end point phosphorus content was ±0. 006%