This study examined instructors' spontaneous responses to student errors in a scripted Tier 2 reading intervention and the relationship between tutor responses and student comprehension outcomes. A sequential exploratory mixed-methods design was used to identify the types of behaviors tutors exhibited in response to student errors, using transcripts of lessons. Tutors used four types of off-script behaviors when reacting to students: (a) scaffolding, (b) telling, (c) unclear feedback, and (d) erroneous feedback. Differences in how tutors implemented the standard protocol were analyzed qualitatively and described using frequency counts. Tutors exhibited differences in the frequencies of each of the behaviors, and differed in how closely they adhered to scripted lesson. Although tutors overall exhibited 76.3% fidelity of implementation, certain components of the lessons were frequently omitted -modeling of the strategy, describing the purpose of the strategy, and providing opportunities for practice. These omissions may have influenced overall responsiveness for students receiving intervention. To determine how tutor differences might influence student outcomes, the frequency counts of the four spontaneous tutor behaviors were entered into regression equations to predict posttest scores on three measures of reading comprehension -Maze, Gates MacGinitie, and ASKIT. Findings indicate that scaffolding was related to student growth on one curriculum based measure of reading comprehension. The other three behaviors -telling, unclear feedback, and erroneous feedback -were not significantly related to student outcomes. Limitations, in light of these findings are considered. Implications for planning intervention studies and tutor training, as well as future directions for research, are discussed.