Evaluating a student’s ability to learn rather than what he or she has learned is at the heart of dynamic testing. Such testing has revealed developing expertise in students that static conventional testing has not. However, low achieving students show more gains from pre to post-tests than high achieving students. Although it seems obvious that poorer students need more help than stronger ones, it is less clear what types of assessor-student interactions allow poorer students to progress. In the project « Dynamic Instruction for and Assessment of Developing Expertise », run by the PACE Center at Yale University (2001-2004), we studied assessor-student interactions (Lund, 2005) during scripted individual assessment within a 4th grade math curriculum based on analytical, practical and creative skills (Sternberg, 1997) and using dynamic instruction (Jeltova, 2004). We qualitatively analyzed 12 assessor-student interactions with both low and high achieving students across two mathematical units: 6 interactions in measurement and 6 in geometry. Low and high achievers were selected based on previous test scores where the former had lower scores. Analysis of the nature of assessor-student interactions compared with students’scores on pre and post-tests showed that the assessors that deviated from the scripted dynamic assessment and who were more reactive to students were those whose students progressed the most.