PURPOSE: Three phase III studies (304 [ClinicalTrials.gov identifier: NCT00699972], 305 [NCT00699582], 306 [NCT00700310]) evaluated perampanel, an alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor antagonist, as adjunctive therapy for refractory partial seizures. We report post hoc analyses of pooled study data by randomized dose. METHODS: Patients with partial seizures despite receiving 1-3 antiepileptic drugs were randomized to once-daily placebo, perampanel 8 or 12 mg (studies 304, 305), or placebo, perampanel 2, 4, or 8 mg (study 306). Studies included a 6-week baseline period and double-blind treatment phase (6-week titration; 13-week maintenance). Primary end points were median change in partial seizure frequency (baseline vs. double-blind phase) and percentage of patients achieving >/= 50% reduction in seizure frequency (baseline vs. maintenance). Here, these end points, together with secondary, exploratory, and safety end points, were assessed using pooled study data. KEY FINDINGS: The pooled intent-to-treat analysis set (randomized, treated patients with any seizure data) included 1,478 patients. Median changes in partial seizure frequency were greater with perampanel than placebo (perampanel 4 mg, -23.3%; 8 mg, -28.8%; 12 mg, -27.2%; placebo, -12.8%; p < 0.01, each dose vs. placebo), as were 50% responder rates (perampanel 4 mg, 28.5%; 8 mg, 35.3%; 12 mg, 35.0%; placebo, 19.3%; p < 0.05, each dose vs. placebo). In addition, median changes in complex partial plus secondary generalized seizure frequency were also greater with perampanel than placebo (perampanel 4 mg, -31.2%; 8 mg, -35.6%; 12 mg, -28.6%; placebo, -13.9%). Perampanel was generally well tolerated. The most frequent treatment-emergent adverse events (TEAEs) were dizziness, somnolence, and headache. Most TEAEs were mild/moderate; relatively few patients experienced severe TEAEs (placebo, 5.4%; perampanel, 8.9%) or serious TEAEs (placebo, 5.0%; perampanel, 5.5%). There were no deaths and no clinically important mean changes in laboratory values, electrocardiography (ECG) findings, or vital signs. SIGNIFICANCE: Perampanel reduced partial seizure frequency and improved responder rates compared with placebo, with an acceptable tolerability profile.