Abstract. The natural and surrogate key architectures are two competing approaches to specifying keys in relational databases. We analyze the performance of these approaches with respect to the update complexity and the query data and expression complexity. Our results provide new insights into the advantages and disadvantages of both approaches.