Suppose you have a data set that represents a mixture of two different concepts, for example, orders and products. The retrieval key is the order number. If order number 1 includes a toaster and order number 15 includes a toaster, then the information about toasters is in (at least) two different records. The update anomaly is that if you update information about toasters in one record, you must update the other records also to avoid a discrepancy. This extra updating is inefficient, unnecessarily difficult, and prone to errors.