In this paper, we present a learning system with a Semantic Spectrum Analyzer to realize appropriate and sharp semantic vector spaces for semantic associative search. In semantic associative search systems, a learning system is essentially required to obtain semantically related and appropriate information from multimedia databases. We propose a new learning algorithm with a Semantic Spectrum Analyzer for the semantic associative search. A Semantic Spectrum Analyzer is essential for adapting retrieval results according to individual variation and for improving accuracy of the retrieval results. This learning algorithm is applied to adjust retrieval results to keywords and retrieval-candidate data. The Semantic Spectrum Analyzer makes it possible to extract semantically related and appropriate information for adjusting the initial positions of semantic vectors to the positions adapting to the individual query requirements.