This paper concerns sequential decision problems with expected utility criteria. A class of such problems, which combines and extends the standard cases of transient, positive and negative dynamic programming, is proposed and analyzed. The principal results are characterizations of when conserving and unimprovable strategies are optimal.