This report investigates seven document representations--configurations of controlled and free-text vocabulary--which can be used to search the INSPEC (Computer and Control Abstracts) and PsychInfo (Psychological Abstracts) databases. The performance of each representation is analyzed, as is overlap among the representations, i.e., the extent to which the same documents are retrieved when searching with different vocabulary configurations. The study's use of a DIALOG simulator known as DIATOM, the participation of 7 trained searching intermediaries, and the soliciting of search questions from 114 online users are described. Major differences between the two databases in terms of which representations perform most effectively, and consistently low overlaps among representations are reported. Results are also discussed in terms of the cumulative improvement on retrieval performance as representations are added sequentially. A probabilistic model of overlap is developed based on the assumption of random retrieval, and this model is fitted against the obtained asymmetric overlaps and the incremental improvements obtained by different overlaps. A total of 20 tables and 15 references are provided. Appendices comprise intermediary training materials, instructions to study participants regarding citation relevance judgements, directions to online users, and sample forms for searchers, as well as the study's Latin square and factorial design, analysis of variance summary results, and theoretical model proofs. (Author/ESR)