THE ABC (APPROACH BY CONCEPT) SYSTEM FOR STORAGE AND RETRIEVAL OF SCIENTIFIC AND TECHNICAL INFORMATION HAS BEEN SUBJECTED TO AN EXTENSIVE PERFORMANCE TEST. IN LI 000 318 THE TEST PROGRAM AND SETUP WAS DESCRIBED AS WELL AS THE METHODOLOGY, AND IN THE SECOND REPORT, LI 000 317, A PRELIMINARY STATISTICAL EVALUATION OF THE TEST RESULTS WAS GIVEN. THIS REPORT DISCUSSES A METHOD OF CONSTRUCTING A MATHEMATICAL MODEL FOR THE FIRST TEST OF THE SYSTEM AND CALCULATES 95 PERCENT CONFIDENCE INTERVALS FOR THE RELEVANCE AND RECALL VALUES. A STATISTICAL, PROBABLISTIC MODEL WAS CHOSEN FOR THE ANALYSIS OF THE TEST DATA, INCLUDING ESTIMATION OF PRECISION AND RECALL AND DETERMINATION OF THE CONFIDENCE INTERVALS. THE BASIC ASSUMPTIONS OF THE STATISTICAL ANALYSIS, NAMELY, THE EXISTENCE OF A CONSTANT PRECISION PARAMETER P AND THE EQUIVALENCE OF THE RETRIEVAL PROCESS UNDER TEST CONDITIONS WITH A POISSEN PROCESS, ARE SATISFIED INDICATING THE SOUNDNESS OF THE MODEL. IT WAS ALSO FOUND THAT RECALL, THE EXPECTED VALUE OF THE AVERAGE RECALL RATIO, IS HIGH WHEN ESTIMATED FROM SAMPLES OF DATA WITH FEW RELEVANT DOCUMENTS. AS THE NUMBER OF RELEVANT DOCUMENTS INCREASES, RECALL SHOWS A DECREASE. THESE PREDICTIONS ARE VERIFIED BY THE ACTUAL TEST RESULTS. IN THE APPENDICES TO THE REPORT, THE VARIOUS NECESSARY FORMULA FOR THE MODEL ARE DERIVED. AN ANNOTATED BIBLIOGRAPHY OF TWENTY-EIGHT ITEMS IS ALSO GIVEN. THIS DOCUMENT IS AVAILABLE AS AD-658-668 FROM THE CLEARINGHOUSE FOR FEDERAL SCIENTIFIC AND TECHNICAL INFORMATION, SPRINGFIELD, VIRGINIA 22151, $3.00 FOR HARD COPY, $0.65 FOR MICROFICHE, 39 PAGES. (AUTHOR/CM)