Cycle counting is a recently devised procedure for continuously updating inventory record accuracy as a replacement for an annual complete inventory count. Cycle counting has been shown to provide greater inventory record accuracy for financial control and for production planning and purchasing operations. However, cycle counting procedures have used an arbitrary basis for classifying inventoried items and for setting the count frequency within each class. In this paper, cycle counting is characterized as stratified statistical sampling. Procedures are shown for maintaining the statistical accuracy required for financial control while reducing the amount of sampling and counting costs. Other criteria can be similarly used. These procedures include the Tschuprow-Neyman optimum allocation rule, the Dalenius-Hodges strata identification rule, a search for an optimum number of strata and improvements in the cycle counting procedures and crew size toward an optimum level: compatible methodologies. A numerical example of typical industrial conditions is provided to show perspective improvements using the optimal stratified cycle counting system in contrast to current cycle counting practices.