A simple model of memory load based on the order that information is searched in is introduced, and minimum memory load search orders are characterized. When tested in an experiment the model is found to track subjects’ choice errors exceptionally well. Together, these results indicate that in the type of rich search problems that are common in modern economic life, optimal sequential search is not feasible for humans because of the excessive memory load it requires. Nevertheless, that the model of memory load is found to explain subjects’ main systematic deviations from optimality suggests that a model of optimal sequential search subject to memory limitations could describe search behavior well. In addition, the model of memory load gives a structural explanation for previously elusive concepts such as choice error, information overload, and how heuristics are selected. Further, it can explain a variety of previously puzzling phenomena observed in rich search experiments, and offers a fully operational exclusion criterion of the type called for in the extended choice literature.