The water/wastewater allocation problem has been widely and efficiently addressed to minimize the freshwater consume using mathematical programming. The biggest challenge on the mathematical procedures is the presence of non-linearities. Aside from stochastic approaches (Genetic algorithms and other), which despite their efficiency, do not guarantee global optimality, many mathematical programming approaches using LP, NLP, MILP, and MINLP were developed (Takama et al., 1980; El-Halwagi and Manousiouthakis, 1990; Galan and Grossmann, 1998; Alva-Argaez et al., 1998; Bagajewicz et al. 1999; Bagajewicz and Savelski, 2001; Karappiah and Grossman, 2005) and their reliability and guarantee of optimality are always a trade off. For the case where the water using units are handled as mass exchangers and single contaminant, the methodologies involving linear programming use the optimality conditions proposed by Bagajewicz and Savelski (2000). These necessary optimality conditions allow transforming non-linear models into linear ones. However, this optimality condition is only valid when the objective function is to minimize the freshwater consumption. In this paper we intend to verify the effects of the optimality condition on problems involving costs. A MILP and a MINLP model that optimize cost are presented and compared. The results prove that the necessary optimality conditions (every process at its maximum outlet pollutant concentration) cannot be used to optimize costs and it can generate significant differences between the found optimum solutions.