Questions on the Scholastic Aptitude Test (SAT) with the largest score differences between women and men of all racial and ethnic groups were identified. Patterns of difficulty that would explain the SAT's continuing underprediction of female first-year college performance were studied. An item analysis of one form of the June 1986 SAT for 1,112 students identified 17 questions with large sex differences. A subsequent item analysis of the November 1987 SAT used the responses of 100,000 high school seniors to identify 23 questions with substantial differences in the numbers of men and women who answered them correctly; 17 of these questions were in the mathematics section. African-American women exhibited the smallest gender gap and Hispanic women the largest when compared with men within their own racial and ethnic groups. The findings of this and other studies suggest that the test publisher could take steps to address the SAT's underprediction and bias. These are summarized as: (1) elimination of questions of identified bias; (2) testing a more balanced array of skills and knowledge; (3) publicity for validity studies about performance prediction; (4) more research on the correlation between SAT scores and grades; and (5) more test-taking time. Thirty-two supporting tables and nine appendices presenting aspects of the study are provided. A 219-item list of references is included. (SLD)