This pilot study examines whether learning without errors is advantageous compared to trial-and-error learning in people with dementia using a procedural task and a randomized case-control design. A sample of 60 people was recruited, consisting of 20 patients with severe dementia, 20 patients with mild-to-moderate dementia and 20 participants without dementia. The participants had to acquire a novel procedural problem-solving task with the help of cues (errorless learning) or with cues only given in case an error was made (trial-and-error learning). The number of steps completed without assistance immediately after errorless or trial-and-error learning and after a delay of 1-3 days was recorded. Overall performance was better after errorless learning compared to trial-and-error learning (p=0.012), with effect sizes being largest in the mild-to-moderate dementia group after delayed testing (d=1.61). The effects of errorless learning were larger after delayed testing. We conclude that errorless learning of a procedural task results in a better performance than learning with errors. This study is the first to use a controlled group design to show such a benefit in patients with dementia. These findings can be extended to the acquisition of tasks that are relevant for everyday-life functioning, having important implications for dementia care.