During the last few decades many Arabic question answering systems have been developed. These systems may extract answers from texts or web-pages. None of these systems make use of question answering database where user can present questions in natural language which differ from the stored questions. The proposed system uses information retrieval approaches to get to the closest answers to the input question, so the system gives partially or totally correct answers. The Latent Semantic Indexing (LSI) is implemented to enhance the resultant selections. Arabic natural language processing is used in the proposed system along with LSI.Keywords-question answering system; natural language processing; information retrieval; latent semantic indexing I.