On-line analytical processing (OLAP) systems deal with analytical tasks in businesses. As these tasks do not depend on the latest updates by transactions, it is assumed that the data used in OLAP systems are kept in a data warehouse, which separates the input coming from operational databases from the output going to dialogue interfaces for OLAP. In this article we present a 3-tier architecture for data warehouses and OLAP systems capturing the fundamental requirement of separating input from operational databases from output to OLAP systems. On this basis we start developing refinement rules to enable step-wise refinement for such systems, which includes pragmatic guidelines for the application of such rules.