In the United States, school-based mentoring programs are a large and widely funded form of mentoring. Despite widespread support, meta-analyses indicate that the effects of school-based mentoring programs are small. One hypothesis for these results is that school-based mentors are not able to develop a sufficiently high-quality relationship with mentees to produce the hypothesized positive effects. This study presents a reanalysis of a large randomized controlled trial of school-based mentoring and examines the estimated effect of mentoring as a function of mentee-reported relationship quality using a novel statistical approach. Although we found that average effect sizes were near zero and consistent with researchers' original findings, our findings also indicate that low relationship quality is associated with some harmful effects, particularly on misconduct, and that as relationship quality improves, so do effects. However, we found that this association decelerates and resulted in small, positive effects for some outcomes. These results suggest that that poor relationship quality may produce harmful effects and a strong relationship may not be sufficient to produce moderate, positive academic and behavioral outcomes in a school-based mentoring context.