We sought to determine the effects of 12 months of power training on cognition, and whether improvements in body composition, muscle strength, and/or aerobic capacity (VO2peak) were associated with improvements in cognition in older adults with type 2 diabetes (T2D). Participants with T2D were randomized to power training or low-intensity sham exercise control condition, 3 days per week for 12 months. Cognitive outcomes included memory, attention/speed, executive function, and global cognition. Other relevant outcomes included VO2peak, strength, and whole body and regional body composition. One hundred and three adults with T2D (mean age 67.9 years; standard deviation [SD] 5.9; 50.5% women) were enrolled and analyzed. Unexpectedly, there was a nearly significant improvement in global cognition (p = .05) in the sham group relative to power training, although both groups improved over time (p < .01). There were significant interactions between group allocation and body composition or muscle strength in the models predicting cognitive changes. Therefore, after stratifying by group allocation, improvements in immediate memory were associated with increases in relative skeletal muscle mass (r = 0.38, p = .03), reductions in relative body fat (r = −0.40, p = .02), and increases in knee extension strength were directly related to changes in executive function (r = −0.41, p = .02) within the power training group. None of these relationships were present in the sham group (p > .05). Although power training did not significantly improve cognition compared to low-intensity exercise control, improvements in cognitive function in older adults were associated with hypothesized improvements in body composition and strength after power training.