The aim of this thesis is to compare the effectivity of using a natural language and a controlled vocabulary when performing subject searches in the bibliographic database LISA Library and Information Science Abstracts. 20 topics were created and for each topic two searches were made, one by using a query formulated with natural language directed to the abstract field and the other one by using a query formulated with descriptors from the thesaurus in LISA directed to the descriptor field. The average relative recall amounted to 64,8% for the AB-queries and 54% for the DE-queries. Failure analysis showed that the majority of the relevant documents not retrieved by AB-queries could have been retrieved with other query terms, such as synonyms. In a natural language there are many different ways of expressing a topic, and this tends to have a negative effect on retrieval effectivity. Very few of the relevant documents would have been retrieved through improved use of search options, such as truncation, and only in a few cases we found that documents were not retrieved because of poor topic description in the abstracts. Failure analysis of relevant documents not retrieved by DE-queries showed that half the documents could have been retrieved if different descriptors had been employed. The other half was not retrieved because of poor topic description. The latter result demonstrates some of the known problems of manual indexing with a controlled vocabulary; the subjectivity of the indexer and the lack of consistency which might appear with regards to term specificity and exhaustiveness.