The present study sought to investigate the use of prediction during sentence comprehension and whether prediction could lead to false memories for sentences when the ending of a sentence is not properly recovered. Participants listened to sentences (e.g. “The boy will eat the cake”) whilst looking at supporting visual scenes. Sentences either contained a constraining verb that could refer to only one object in the scene (e.g. “eat”) or a non-constraining verb that could apply to any object in the scene (e.g. “move”). Some sentences contained a filler word (“thingy” or “something”) instead of the final noun of the sentence. Participants were presented with a memory test after listening to a block of sentences and had to respond whether a visually presented sentence had previously been heard or not. Critical sentences had the filler word replaced with the correct noun to test for false recognition of sentences with predicted words. Results showed that participants looked towards a target object before hearing its name significantly more and earlier when the sentence heard contained a constraining verb than when the sentence contained a non-constraining verb. Greater looks towards distractor objects in the scene were found in the non-constraining verb condition than the constraining verb condition. In the memory tests, participants were significantly worse at rejecting critical sentences than rejecting new sentences or accepting sentences previously heard. However, there was no significant difference between performance on critical trials in which participants had engaged in predictive eye-movements and critical trials in which participants had not engaged in predictive eye-movements. Results provide further support for prediction during online sentence comprehension but the nature of the representation of the predicted item and the consequences of an incorrect or unresolved prediction remains an open question.