The goal of this study was to find a way to quantify three criteria of representational quality, described by Greeno, so that it would be possible to examine statistically the relationship between representational quality and other variables related to problem solution. The sample consisted of 18 college students, 84 percent of whom had mathematics through trigonometry or calculus in high school as well as at least one college mathematics course. Students were asked individually to think aloud as they solved six algebra word problems, and then to give a retrospective report describing what could be remembered about the problem-solving process used. Each idea unit in the protocols was coded, using a slightly modified coding system developed by Ericsson and Simon. Then, coherence and correspondence scores were computed for each problem. Both scores were found to predict whether or not a student would be able to solve correctly at least some word problems, with the correspondence score a stronger predictor. Three types of errors made by students were also noted: interpretive, integrative, and recall errors. Appendices contain the word problem set, a sample coded verbal protocol, and a task analysis for one problem. (MNS)