OBJECTIVES Health plans often conduct analysis of patient adherence to alternative anti-depressant options using only pharmacy data. This study assesses how various factors identified on medical claims and pharmacy data records impact compliance, a common measures of adherence. METHODS Retrospective study of initiators on selective serotonin and serotonin-norepinephrine reuptake inhibitors (SSRIs and SNRIs) during 2005. Study patients (n=251,934) were continuously eligible 6 months before and 12 months after initiation on their index medication (duloxetine, venlafaxine XR, citalopram, escitalopram, fluoxetine, paroxetine, or sertraline). Compliance was defined as percentage of patients with 365-day medical possession ratio ≥ 80%. Analysis was based on chi-square tests and logistic regression. All reported differences were statistically significant at p lt 0.01. Confidence intervals for all reported odds ratios were significant at the 95% level. RESULTS Across all SNRI patients, compliance was the same for venlafaxine XR and duloxetine (36%). Among SNRI patients with diagnosed MDD, however, compliance was greater for duloxetine (44%) than venlafaxine XR (41%). Across all SSRI patients, compliance was highest for citalopram (29%), lowest for escitalopram (26%). Among SSRI patients diagnosed with MDD, however, compliance was highest for those on paroxetine (33%), lowest on citalopram (29%). Significant predictors of compliance based on pharmacy claims data among MDD patients included prior use of antidepressants (OR=1.72) and initial dosing at sub- or super-therapeutic levels (OR=0.85 and 0.92, respectively). Significant predictors based on medical claims data included recent visits to a mental health specialist (OR=1.21), a co-morbid diagnosis of GAD (1.20), and the presence of co-morbid baseline pain (OR=.85). CONCLUSIONS Measures of medication compliance may be impacted by a number of factors, including patient characteristics, medical history, and physician specialty. Adherence analysis should adjust for relevant confounders identified in medical claims, as well as pharmacy data records.