Abstract: The rapid development of network technology and the semantic web has hastened the development of spatial information sharing and application services. However，the functions provided by the existing spatial information service models still cannot meet the needs of some complex applicadons，and how to achieve a combination of on-demand services is becoming a hot research issue. Currently，multiple services are combined in a semi-automatic or automatic manner to meet various applicanon requirements but low time efficiency remains a common problem.This study surveyed the advantages and disadvantages of related service composition methods，and developed an intelligent approach based on neural network theory to correctly and efficiently combine some commonly used spatial information processing services. This method adopts an extended ontology representation model for ad hoc services to provide semantic description，and utilizes the principle of synaptic neural net-work for searching core services. Finally，service composition is realized by an integrated indexing function. Experiments show that the method not only effectively improved the computation efficiency，but also lowered the barriers for intelligent service composition for non-professionals，confirming its practical value in handling various application requirements.