Overdose Education and Naloxone Distribution (OEND) is an effective public health intervention to reduce opioid overdose fatalities (McDonald and Strang, Addiction 111:1177–1187, 2016). However, we know little about OEND implementation outcomes (i.e., indicators of implementation success), specifically the fidelity of training delivery, and how these may relate to intervention outcomes (i.e., indicators of the success or effectiveness of an intervention), such as overdose knowledge and attitudes. This study evaluated 16 OEND trainings conducted at different Opioid Overdose Prevention Programs in New York City. Trainees (N = 75) completed the Opioid Overdose Knowledge and Attitude Scales before and after training (intervention outcomes). Implementation outcomes were fidelity (competence and adherence of the trainer, N = 10; modified Fidelity Checklist) and acceptability of OEND (Acceptability of Intervention Measure), assessed from multiple perspectives (trainees, trainers, and an independent observer). Trainees’ overdose knowledge, t(71) = − 8.12, p < 0.001, 95% CI [− 6.54, − 3.96], and attitudes, t(65) = − 6.85, p < 0.001, 95% CI [− 0.59, − 0.33], improved significantly from pre- to post-training. Stepwise multiple regression models indicated that adherence of the trainer rated from the observer perspective added significantly to the prediction of changes in overdose knowledge, F(1, 67) = 9.81, p = 0.003, and explained 13% of the variance in outcome. However, fidelity measures from the perspective of trainees or trainers and acceptability of OEND were not associated with changes in trainees’ overdose knowledge or attitudes. OEND implementation outcomes and their relationship with intervention outcomes differed depending on the role of the fidelity rater in relation to the intervention. Specifically, our findings indicate that fidelity should be measured from an independent perspective (i.e., an individual who is experienced with fidelity rating but not directly involved in the intervention).