A coding scheme is presented and used to evaluate solutions of seventeen students working on twenty five stoichiometry problems in a think-aloud protocol. The stoichiometry problems are evaluated as a series of sub-problems (e.g., empirical formulas, mass percent, or balancing chemical equations), and the coding scheme was used to categorize each sub-problem solution as successful, neutral, or unsuccessful, with more detailed codes comprising the neutral and unsuccessful categories, for a total of eight codes. A relatively high frequency of neutral results was found in which students simply did not realize when or how to approach a sub-problem. A lack of conceptual understanding of the mole concept appears to be closely related to students skipping crucial steps in stoichiometry problems, especially the sub-problems stoichiometric ratio and mole concept. Students' failures were also observed to be due to a lack of basic knowledge, such as the names of chemical compounds. The application of the new code system was shown to reveal difficulties that might have otherwise been missed by an analysis that focused on end results only.