Since the inception of ontology systems, we have witnessed both the number of ontologies appearing on the web growing rapidly and their size expanding dramatically. The increase in the number of ontologies brings problems in querying multiple ontologies that are possibly heterogeneous and independently developed, and the expansion of ontology size poses challenges in system efficiency and management. While much attention has been paid to ontology integration systems so as to enable interoperability between different ontologies, the problem of how to handle ontologies of large size has not been sufficiently addressed. In this thesis, we describe a distributed ontology system (DOS) which is more reliable and scalable than centralized ontology systems where ontologies are stored on a single computer. The idea of DOS is to divide a large ontology into smaller pieces,called ontology fragments, then distribute these fragments over a set of autonomous nodes where each of them is equipped with a reasoner. We formally define DOS by presenting a logical model for DOS and discuss its advantages and challenges in terms of query processing and collaborative ontology development. In addition, we describe a query processing procedure for conjunctive queries with ontologies that have large number of instance. We prove that our method always returns sound answers for OWL ontologies, and returns complete answers for SHI ontologies. Furthermore, using the procedure we show how to handle two other reasoning problems, consistency checking introduced by updating ABox and instance realization, in DOS. Finally, we discuss how DOS provides synchronous and nonlocking editing for collaborative ontology development.