In this paper the risk-sensitive control of parially observed Markov decision processes is considered. The replacement problem is analyzed in this context, and the structure of risk sensitive optimal controllers is given.