In this paper we study the convergence properties of the power series algorithm, which is a general method to determine (functions of) stationary distributions of Markov chains. We show that normalization plays a crucial role and that the convergence can be improved by introducing some minor changes in the algorithm. We illustrate this with several numerical examples. Key words and phrases. Coates graphs, Markov chains, power series, stationary probabilities AMS 1991 Subject classifications. Primary 60K25; Secondary 60J25