Evaluations in the field of police studies do not adequately address implementation fidelity, which impacts understanding of outcomes, how and why an intervention works in practice, what can be done to achieve improvements, and how to implement effective innovations in different contexts. Responding to calls for greater use of implementation science to improve innovations in policing, this article draws on the qualitative phase of a realist, quasi-experimental evaluation to examine implementation fidelity of police drug diversion (PDD) schemes in three police forces across England. The research principally involved interviews and focus groups with 221 people who were designers, deliverers, and recipients of PDD. Quantitative data on the recorded outcomes of those eligible for diversion was analysed to assess police use of PDD schemes. Findings on adherence indicate that PDD had lower than intended coverage owing to the exercise of discretion by frontline police officers. Moderators of fidelity are examined to further understanding of the realities of implementation processes, explain why officers were not diverting eligible offenders, and assess attempts by police managers to constrain discretion. The analysis demonstrates the complex interactions between moderators. It is structured around intervention complexity, facilitation strategies, and quality of delivery, but also covers participant responsiveness and capacity, organisational and cultural context, and time. The conclusion stresses the need for evaluations to consider implementation fidelity and makes recommendations for improving fidelity when implementing PDD schemes.