Although the efficacy of small-group, risk reduction interventions based on cognitive behavioral principles has been widely documented in HIV behavioral research literature, little is known about how AIDS service organizations (ASOs) view these research-based models. From a nationwide sample of 77 ASOs, this study assessed factors influencing attitudes of prevention program directors and frontline staff toward research-based interventions. Characteristics of individual respondents as well as organizational characteristics of the ASO itself were used to predict perceived benefits of adopting this type of intervention, perceived efficacy (confidence) in the ASO's ability to implement it, and perceived barriers to adoption. Findings revealed uniformly positive perceptions of benefits among respondents from ASOs of different sizes and organizational experiences, although directors held more favorable evaluations than frontline staff. Respondents from ASOs that were larger, had previously delivered group or workshop interventions, or had received outside technical assistance in the past expressed more confidence in the ability of their ASO to implement the intervention. On the other hand, older and more highly education individuals had less confidence in their organization's ability to implement the model. Resource constraints (money, staff, and time) were the most common barriers cited by the respondents. Overall, higher organizational role and longer tenure at an ASO were associated with the perception of more barriers to adopting science-based interventions. Respondents from ASOs with a history of receiving technical assistance reported fewer perceived barriers. The successful dissemination of HIV prevention models from the research arena to the service arena will require mechanisms to provide appropriate funding and technical assistance, particularly to smaller organizations. Mindful of the resource constraints faced by ASOs, researchers can facilitate this process by attempting to develop interventions that are less resource- and time-consuming than current models.