The use of computers to build diagnostic inferences is explored in two contexts. In computerized monitoring of liquid oxygen systems for the space shuttle, diagnoses are exact because they can be derived within a world which is closed. In computerized classroom testing of reading comprehension, programs deliver a constrained form of adaptive testing and error performance summary. However, the world is open: diagnostic inferences cannot be made with precision, and additional practical factors play an important role in delimiting the usefulness of such a system. Problems of uncertainty, negation, and nondeterministic prediction are also discussed.