Dynamic lot sizing heuristics commonly implemented in standard MRP software perform well under conditions of systematic demand only. A survey and numerical test of methods proposed in the literature shows that techniques are available which cope efficiently with both systematic and erratic demand. The paper discusses theoretical and practical ramifications of the dynamic lot sizing problem, surveys heuristics published in the U.S. and West German literature, reports summarily on the findings of extensive simulation experiments involving these heuristics, and, as a conclusion from analytical and numerical results, proposes an algorithm which enhances performance under conditions of erratic demand.