The degree to which performance assessments meet their dual mandate to evaluate student learning and inform instructional practice is not adequately addressed through traditional concerns for reliability and validity. A possible approach is suggested for examining the cognitive activity students engage in during a performance assessment. The approach is demonstrated with the "Mystery Powders" classroom-based assessment being piloted by several large school districts. Thirty-seven fourth- and fifth- grade students were interviewed while they conducted an investigation to determine properties of various powders. Interview protocols and observations were analyzed, and high and low scorers were described. Results indicate that although performance scores and general understanding were generally low, high scorers could be distinguished on several characteristics. Results support the viability of the approach for analyzing the extent to which performance assessments measure higher-order thinking. Implications for instructional practice are considered. Seven figures and four tables are included. (Contains 7 references.) (Author/SLD)