In aluminium smelters, the operational staffs constantly face decision making situations for operation and process control and these decisions can have significant impact on the process. The smelting process involves highly complex mechanisms and has rich information but low observability. In this environment, without support tools, systematic information management, or robust control models, decision making is challenging. This paper discusses different types of decision making processes and demonstrates that naturalistic decision making models (Recognition-Primed Decision Making, ie RPD) are more suitable to describe the situations in smelters. A model which combines an advanced control model, a system and human interactive approach and the thinking process in RPD is proposed to improve the quality of decisions for the operational staffs in smelters, hence the efficiency and productivity of the process.