The construct validity of algebra word problems for measuring quantitative reasoning was examined, focusing on an analysis of problem attributes and on the analysis of constructed-response solutions. Constructed-response solutions to 20 problems from the Graduate Record Examinations (GRE) General Test were collected from 51 undergraduates. Regression analyses indicated that models including factors such as the need to apply algebraic concepts, problem complexity, and problem content could account for 37% to 62% of the variance in problem difficulty. Four classes of strategies were identified for constructed response problems: equation formulation, ratio setup, simulation, and other (unsystematic) approaches. Higher achieving students used equation strategies more and unsystematic approaches less than lower achieving examinees. Problem conception errors were the best predictor of performance on the constructed-response problems and the Scholastic Aptitude Test mathematics test (SAT-M). In contrast, procedural errors contributed to the prediction of performance on the constructed-response problems but not to standing on the SAT-M. Overall, these results provide support for the construct validity of GRE algebra word problems and of SAT-M as measures of quantitative reasoning. Appendix A provides some sample problems, and Appendix B is a table of attribute codings. (Contains 10 tables, 1 figure, and 41 references.) (Author/SLD)