This study is based on data from a state-wide assessment that included both multiple-choice and constructed-response items. The intent of the study was to see whether item types make a difference in gender results. The items on both tests were categorized according to whether they assessed procedural knowledge, concepts, problem solving, or mathematical communication. Data came from Delaware, which in 1995 administered two tests in mathematics to all public school students in grades 3, 5, 8, and 10. A total of 29,809 students were tested on the Delaware "Interim Assessment," a constructed-response test, and the Iowa Tests of Basic Skills (ITBS) (or the ITBS Tests of Achievement and Proficiency Survey Battery in grade 10). On the Interim Assessment in grade 3, the female mean raw score was higher, but at all other grades, males outperformed females. On the ITBS, males scored higher in grades 3 and 8, and there was no difference at grade 5 or 10, although the male mean score was higher at every grade. Differences generated by the broad and narrow interpretations of problem solving show some gender differences that might not have been apparent otherwise. Overall, males outperformed females on problem solving at grades 8 and 10. Results in some ways contradict the more hopeful conclusions of other studies that have shown the gender gap to be narrowing, though they do affirm some results of a study that showed males stronger in problem solving in the high school years. Results suggest that, while the gap is narrowing on traditional multiple choice tests, it is still present on more complex items that require students to construct their own responses and communicate their thinking. It is especially disturbing to see that the gap increases with grade level, which is in keeping with other studies showing females falling behind in adolescence. An appendix contains the item characterization protocol of the 1995 Delaware Mathematics Assessment. (Contains 6 tables and 16 references.) (SLD)