Abstract There are 4 sources of bias in clinical trials: investigator bias, patient expectation (placebo response), ascertainment bias (inadvertent selection of an unrepresentative sample), and nonspecific effects such as the normal waxing and waning of symptoms over time and the quality of the doctor-patient relationship. In drug trials, these biases are adequately controlled by comparing active to inert pills, randomly assigning subjects to treatments, blinding both the investigator and subject to group assignment, and testing subjects at multiple sites. However, there are special problems with conducting clinical trials of behavioral or psychological interventions that render these controls inadequate. It is impossible to blind the experimenter to which treatment is active, it is difficult to identify a control treatment that is inactive but just as credible to the subject, and doctor-patient relationship variables are more important than in drug trials. The inability to blind the experimenter can be circumvented by having an independent, blinded investigator assess the outcome, and doctor-patient effects can be controlled by using multiple, experienced therapists. The most difficult problem, identifying an appropriate control treatment, can be solved by adhering to 2 principles: the control treatment should be plausible, and it should not have a significant impact on the mechanism that is thought to explain the effectiveness of the investigational treatment. Investigators should confirm that these 2 goals have been achieved by monitoring expectation of benefit with a treatment credibility questionnaire, measuring changes in process variables (variables that reflect the presumed mechanism of treatment), and monitoring differential dropout rates.