In order to investigate features of the misinformation effect, sixty-five participants viewed a 9-minute video (Day 1), then 24 hours later read a narrative description of the video events (Day 2). Narrative descriptions were manipulated by replacing 20 original items appearing in the video with either misinformation (semantically or non-semantically related to the original) or no misinformation (a superordinate alternative or omission). Twenty-four hours later, participants undertook either a recall test or a recognition test (Day 3). Semantic and non-semantic misinformation (Day 2) both reduced response accuracy (Day 3). Whereas semantic misinformation was more likely to be reproduced on Day 3 (misinformation interference), non-semantic misinformation was more likely to produce other responses (misinformation acceptance). Recognition testing produced greater accuracy, but was associated with more misinformation interference, while recall testing produced more misinformation acceptance. These results arguably provide support for independent contributions by episodic and semantic memories to erroneous memory reports.