Mild-to-moderate AD patients were randomized to placebo or rosiglitazone (RSG) 2, 4 or 8 mg. Primary end points at Week 24 were mean change from baseline in AD Assessment Scale-Cognitive (ADAS-Cog) and Clinician's Interview-Based Impression of Change Plus Caregiver Input global scores in the intention-to-treat population (N=511), and results were also stratified by apolipoprotein E (APOE) genotype (n=323). No statistically significant differences on primary end points were detected between placebo and any RSG dose. There was a significant interaction between APOE epsilon4 allele status and ADAS-Cog (P=0.014). Exploratory analyses demonstrated significant improvement in ADAS-Cog in APOE epsilon4-negative patients on 8 mg RSG (P=0.024; not corrected for multiplicity). APOE epsilon4-positive patients did not show improvement and showed a decline at the lowest RSG dose (P=0.012; not corrected for multiplicity). Exploratory analyses suggested that APOE epsilon4 non-carriers exhibited cognitive and functional improvement in response to RSG, whereas APOE epsilon4 allele carriers showed no improvement and some decline was noted. These preliminary findings require confirmation in appropriate clinical studies.