In the 1992 mathematics assessment of the National Assessment of Educational Progress (NAEP), approximately one-third of the questions and about half of the student response time were devoted to questions asking students to construct their own responses. Regular constructed response items required a short answer with a problem solution, while extended constructed response required demonstration of student reasoning. Analysis of student papers shows that most made a conscientious effort to respond, but that performance left much to be desired. On regular constructed response questions, the average percent correct by grade level was 42 percent of grade 4, 53 percent of grade 8, and 40 percent of grade 12. For extended constructed response questions, the average percentages of student producing correct responses were 16 percent at grade 4, 8 percent at grade 8, and 9 percent at grade 12. Results indicate that extended constructed response questions can be successfully used in large-scale assessment and that they do signify student proficiency. The levels of performance on multiple-choice items were higher than for constructed responses. Sixty tables present study findings. A procedural appendix explains study methodology, with an additional seven tables. (SLD)