Meta-level architectures for dynamic control of reasoning processes are quite powerful. In the literature, many applications in reasoning systems modeling complex tasks are described, usually in a procedural manner. In this article we present a semantic framework based on temporal partial logic to describe the dynamics of reasoning behavior. Using these models, the semantics of the behavior of the whole (meta-level) reasoning system can be described by a set of (intended) temporal models. © 2002 Wiley Periodicals, Inc.