The error speed effect, the observation that items with a fast recognition error in a single-item old-new recognition task have a higher probability of an incorrect response in a two-alternatives forced choice task compared to items with a slow recognition error, is mostly taken as evidence that some recognition errors are attributable to misleading memory evidence. Recently, Akan et al. (2023) demonstrated that the error-speed effect only occurs within misses but not within false alarms, which challenges the prediction of several recognition models that predict the error-speed effect for both misses and false alarms. In the present study, we investigate whether false alarm can in general be caused by systematically misleading memory evidence or are alternatively only elicited by random fluctuations in the decision process (diffusion model) or by incorrect guesses (two-low threshold-model). Therefore, we manipulated the similarity between studied items and non-studied items to increase systematically misleading memory evidence within false alarms. Using a sequential sampling plan, we found evidence in favour of the occurrence of the error-speed effect within false alarms, indicating that false alarms, too, can sometimes be elicited by systematically misleading memory evidence.