The goal of this study was to evaluate experimentally systematic instruction compared with trial-and-error learning (conventional instruction) applied to assistive technology for cognition (ATC), in a double-blind, pre-test-post-test, randomised controlled trial. Twenty-nine persons with moderate-severe cognitive impairments due to acquired brain injury (15 in systematic instruction group; 14 in conventional instruction) completed the study. Both groups received 12, 45-minute individual training sessions targeting selected skills on the Palm Tungsten E2 personal digital assistant (PDA). A criterion-based assessment of PDA skills was used to evaluate accuracy, fluency/efficiency, maintenance, and generalisation of skills. There were no significant differences between groups at immediate post-test with regard to accuracy and fluency. However, significant differences emerged at 30-day follow-up in favour of systematic instruction. Furthermore, systematic instruction participants performed significantly better at immediate post-test generalising trained PDA skills when interacting with people other than the instructor. These results demonstrate that systematic instruction applied to ATC results in better skill maintenance and generalisation than trial-and-error learning for individuals with moderate-severe cognitive impairments due to acquired brain injury. Implications, study limitations, and directions for future research are discussed.