An experimental test battery (the Johnson O'Connor Research Foundation battery) designed to measure numerical facility was administered to 1,451 subjects at 12 testing centers across the United States over a 5-month period. Five work samples were included: (1) arithmetic; (2) counting backwards; (3) number reasoning; (4) rule learning; and (5) number series. All five work samples had moderately high internal reliabilities. Factor and item analyses showed that each work sample measured primarily one dimension. These work samples provided information that was not obtained with the standard Johnson O'Connor aptitude test battery. When predictive validity was examined, all five discriminated to some degree among groups of examinees divided according to their college majors. Arithmetic and Counting Backwards were, psychometrically, the purest measures of numerical facility; however, all were considered suitable. Twenty-five tables are presented. (SLD)