OBJECTIVES Naturalistic studies comparing differences in risks across antidepressant agents must take into account factors which influence selection of specific agents and may be associated with outcomes. We examined predictors of antidepressant choice among VA patients treated for depression. METHODS Retrospective cohort study of VA patients with depression diagnoses and a new start of one of the seven most commonly prescribed antidepressant agents between 1 April 1999 and 30 September 2004 (n = 502 179). We examined the relationship between patient and facility characteristics and new starts of bupropion, citalopram, fluoxetine, mirtazapine, paroxetine, sertraline, and venlafaxine. We also examined factors associated with new starts only among patients starting selective serotonin reuptake inhibitors (SSRIs). RESULTS Thirty-three percent of patients starting mirtazapine had at least three outpatient mental health visits in the prior year, compared to <= 24% of patients prescribed other antidepressants. Patients starting mirtazapine were also most likely to have received at least two other psychotropic medications in the prior year. Of the four SSRIs, 40% of the patients receiving sertraline and only 31% of those receiving fluoxetine were 65 years or older. A comorbid anxiety disorder (other than post-traumatic stress disorder) was diagnosed in 21% of paroxetine patients compared with <= 15% of other SSRI patients. CONCLUSION Choice of antidepressant medication for depressed VA patients was associated with important differences in demographic and clinical variables, including psychiatric illness severity, older age, and likelihood of a comorbid anxiety disorder. These findings emphasize the importance of controlling for selection bias when using observational data to compare risks from or effect of mental health treatments. Copyright (C) 2010 John Wiley & Sons, Ltd.