In two experiments involving word-stem completion, an advantage was found for errorless over errorful-learning conditions, for both severely and moderately memory-impaired participants. This advantage did not depend on the implicit/explicit nature of the question asked. Additional tests showed that subsequent recognition of target items was good for both groups, but only in the absence of lures derived from participants' prior errors. Source-memory was shown to be virtually absent in the severely impaired group and only weakly present in the moderately impaired group. This combination of results suggests that preserved implicit memory, in the absence of explicit memory, is sufficient for an errorless-learning advantage to accrue. (c) 2005 Elsevier Ltd. All rights reserved.