We use the ideas developed by Madan and Milne (1994. Mathematical Finance 3, 223-245), Lacoste (1996. Mathematical Finance 6, 197-213) to explore the optimality of polynomial approximations in pricing securities. In particular, we look at the approximations for security payoffs as well as the associated pricing formula in a L2 framework. We apply these ideas to two examples, one where the state variable follows an Ornstein-Uhlenbeck process and one based on Brownian motion with reflecting barriers, to illustrate the strengths and weaknesses of the approach.