A mixed-integer non-linear programming model that minimises the total regulatory costs of controlling nitrogen oxide is used to investigate how a newly proposed permit trading scheme in Taiwan, which incorporates the features of banking and a nonone- to-one trading ratio, may affect firms’ emission reduction strategies and permit trading decisions. Compared to the previous regulation where only an air pollution fee is used, the new regulation that requires a reduction in emissions by 10 per cent from the emission level in the year 2000 for a 5 year period will increase the costs by 77 per cent, which is equivalent to US # 9.87 million. The design of banking and the increasing returns to scale characteristic of pollution control among firms might lead to an uneven reduction in emissions in each year. Setting a lower reservation rate for banking would, however, help maintain a more stable environmental quality without a significant loss to the government in terms of air pollution fee revenue.