The current work presented a computational intelligence approach used for minimizing NOx emissions in a 300 MW dual-furnaces coal-fired utility boiler. The fundamental idea behind this work included NOx emissions characteristics modeling and NOx emissions optimization. First, an objective function aiming at estimating NOx emissions characteristics from nineteen operating parameters of the studied boiler was represented by a support vector regression (SVR) model. Second, four levels of primary air velocities (PA) and six levels of secondary air velocities (SA) were regulated by using particle swarm optimization (PSO) so as to achieve low NOx emissions combustion. To reduce the time demanding, a more flexible stopping condition was used to improve the computational efficiency without the loss of the quality of the optimization results. The results showed that the proposed approach provided an effective way to reduce NOx emissions from 399.7 ppm to 269.3 ppm, which was much better than a genetic algorithm (GA) based method and was slightly better than an ant colony optimization (ACO) based approach reported in the earlier work. The main advantage of PSO was that the computational cost, typical of less than 25 s under a PC system, is much less than those required for ACO. This meant the proposed approach would be more applicable to online and real-time applications for NOx emissions minimization in actual power plant boilers.