This research develops knowledge based computer systems for "qualitative physics," allowing computers to reason more effectively about the physical world. Qualitative physics complements traditional physics by providing (1) fast answers in simple situations and (2) a framework for more detailed analysis. Specific qualitative models are being developed for fluid systems, thermal systems, and motion, adequate to describe complex systems such as as mechanical clocks and boilers. The models include methods for prediction, measurement, interpretation, and for constructing plans and procedures. The research also includes an investigation of how people learn physical domains, including cognitive simulations of analogical processing. The significance of this research is its potential to produce "robust" computer reasoning systems, comparable to humans in their ability to reason approximately about how the physical world around them will behave, without the detailed information and mathematical equations required for classical models of physics. In addition, studying the range of qualitative physics expertise in humans may provide important insight into the nature of cognition. Practical and effective qualitative reasoning systems will also be essential in developing many advanced technological systems: Examples include intelligent monitors for complex systems such as factories and power plants, and intelligent computer assistants for mechanical and engineering design.