In this paper, the authors describe a case of measuring implementation fidelity within an evaluation of Math Recovery (MR), a pullout tutoring program for low-achieving first-graders. They use this case to address two aspects of implementation fidelity studies: (1) their feasibility with respect to unscripted interventions, and (2) their relationship to ongoing program development. The aim of the MR program is to use children's current understandings of number as bases for providing instruction that will support them in constructing increasingly sophisticated strategies. Therefore, assessing fidelity in this case is not as simple as monitoring adherence to a script, but requires assessing the complex practice of delivering mathematics instruction attuned to a child's current understanding and needs. Their intentions were to both measure the extent to which the program was implemented as intended, and link the measures to student outcomes. Determining the extent to which the tutoring is enacted as intended requires an explication of "good" tutoring as defined by the developers and systematically evaluating tutors' practices against that ideal. However, they also go beyond MR's notion of "good" tutoring by looking for instances of "positive infidelity" within tutoring sessions. Thus, they view studies of implementation fidelity as potential sources for refining theory and program design. Appendices are: (1) References; and (2) Tables and Figures.