Objective: Little information is available comparing self- versus investigator ratings of symptoms in adult ADHD. The authors compared the reliability, validity, and utility in a sample of adults with ADHD and also as an index of clinical improvement during treatment of self- and investigator ratings of ADHD symptoms via the Conners Adult ADHD Rating Scale (CAARS). Method: We analyzed data from two double-blind, parallel-design studies of 536 adult ADHD patients, randomized to 10-week treatment with atomoxetine or placebo. Outcome variables included ADHD symptom severity (CAARS self- and investigator ratings), psychiatric symptom comorbidity, and functioning. Results: All five CAARS subscales showed good internal consistency at each time point. Similarly, interrater reliability was acceptable for each subscale. Following treatment, CAARS total scores and subscale scores improved significantly from baseline. CAARS subscales also predicted changes in other psychiatric symptoms and functioning. Overall, baseline investigator ratings were stronger predictors of treatment outcome than baseline self-report scores. Conclusions: The CAARS demonstrated good internal consistency and inter-rater reliability, as well as sensitivity to treatment outcome. The finding of greater predictive power of investigator-rated baseline scores merits further investigation.