Background: Responses to atomoxetine vary for individual patients with attention-deficit/hyperactivity disorder (ADHD). However, we do not know whether any factors can be used to reliably predict how individuals with ADHD will respond to treatment. Objective: Our objective was to evaluate background variables that facilitate early identification of those adults with ADHD who are likely to respond to treatment with atomoxetine. Methods: We pooled data for atomoxetine-treated adults with ADHD from 12 clinical trials for a short-term (10-week) analysis, and from 11 clinical trials for a long-term (24-week) analysis. Patients not meeting a response definition [≥30 % reduction in Conners’ Adult ADHD Rating Scales-Investigator Rated: Screening Version (CAARS-Inv:SV) total score and Clinical Global Impressions of ADHD Severity Scale (CGI-S) score ≤3 at endpoint], or who discontinued, were defined as non-responders. Another definition of response (≥30 % reduction in CAARS-Inv:SV total score at endpoint) was also used in these analyses; only the results with the former definition are shown in this abstract, as the same conclusions were gained with both definitions. A treatment-specified subgroup detection tool (a resampling-based ensemble tree method) was used to identify predictors of response. Results: Of 1945 adults in the long-term analysis, 548 (28.2 %) were responders to atomoxetine at week 24; 65.2 % of 1397 non-responders had discontinued. Of 4524 adults in the short-term analysis, 1490 (32.9 %) were responders at week 10; 33.2 % of 1006 non-responders had discontinued. No analyzed baseline parameters (age, sex, prior stimulant use, ADHD subtype, CAARS-Inv:SV, CGI-S) were statistically significant predictors of response. Reductions in CAARS-Inv:SV total, CAARS-Inv:SV subscores, and CGI-S at week 4 in the short-term analysis, and at weeks 4 or 10 in the long-term analysis, were statistically significant predictors of response, i.e., patients with versus without these reductions early in treatment were more likely to be clinical responders at later time points. Sensitivity ranged from 28.6 to 85.9 %, and specificity ranged from 23.8 to 86.7 %. Predictors with higher sensitivity had lower specificity, and vice versa. Conclusions: Reductions in CAARS-Inv:SV and CGI-S scores at weeks 4 and 10 are statistically significant predictors of response to atomoxetine at later time points in adults with ADHD. However, the predictors identified by these analyses are not reliable enough for use in clinical practice. The only currently available method to judge whether individuals with ADHD will respond to atomoxetine is to start treatment and assess the response over an extended period, sometimes longer than 10 weeks.