• There used to be a "best practice" of 5 to 15 measure groups (fact tables) per cube by the Kimball Group, but that is a very general rule. It has since been backed off by both the Kimball Group and revised by the SQL CAT team.

    It really does depend on how you are going to use the data/uptime requirements/number of scripted calculated members/etc, but here is the experience of a user that had 23 fact tables to deal with in 2005:

    http://www.ssas-info.com/VidasMatelisBlog/40_splitting-analysis-services-2005-cubes-based-on-measure-groups