People use and extend their knowledge of the physical world constantly. Understanding how this fluency is achieved would be an important milestone in understanding human learning and intelligence, as well as a useful guide for constructing machines that learn. This paper presents a theoretical framework that is being developed in an attempt to construct a computational account of human experiential learning in physical domains. Qualitative process theory is used to model portions of physical knowledge, and structure-mapping theory is used to characterize the computations that move a learner from one representation to another. This document outlines both theories and proposes a learning sequence for the physical domains. The stages of understanding which comprise the sequence are: (1) protohistories (prototype histories that serve as summaries of experience); (2) the causal corpus (a collection of causal statements computed from prototype objects and protohistories); (3) naive physics (models which replace causal statements with theories about specific mechanisms of change); and (4) expert models (discovering ways to resolve ambiguities and to construct powerful generalizations). A seven-page reference list is included. (TW)