Ammonia synthesis production is a critical chemical industry around the world. As the key process variable, the ammonia concentration at the ammonia converter outlet reflects the production status and provides good advices for the operators. However, it cannot be easily measured because of high expenditure and deficient reliability of online sensors in a real-world ammonia synthesis process. Due to this, a soft sensor, which is used to predict the outlet ammonia concentration, is developed using BP neural network (BPNN). An improved particle swarm optimization with expansion and constriction operation (PSOEC) is proposed to optimize the weights and thresholds of BPNN. The PSOEC and BPNN based soft-sensing model (PSOEC-NN) is applied to inferring the outlet ammonia concentration in a fertilizer plant. Results using other modeling methods (BPNN and PSO-NN) are presented for comparison purpose. The proposed PSOEC-NN based soft sensor shows high precision and good generalization capability. PSOEC-NN model would offer great help for further work like advanced control and operational optimization in the ammonia synthesis process. (C) 2011 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.