Previous research claiming that there is a metamemory deficit in Alzheimer's Disease (AD) has been based on paradigms in which metamemory judgements are compared with performance. These methods confound predictive accuracy with very poor memory performance. In the experiments presented here this confound is removed by focusing on the sensitivity of metamemory judgements to item differences at encoding, rather than on predictive accuracy. In Experiment 1 participants studied words of high or low recallability, and either made judgements of learning (JOLs) or declared recall readiness. It was found that the AD group discriminate between items in their metamemory judgements to the same extent as age matched controls. Both groups rated the highly recallable words as being more likely to be recalled, and allocated more study time to low recallability items. In Experiment 2 participants were asked to rank the likelihood of recall of items that varied in objective recallability. Once again, AD patients were as sensitive to objective differences in stimuli as controls. Therefore, using measures based on sensitivity to item differences, we find no evidence of a metamemory deficit at encoding in AD. The findings are discussed in terms of metamemory functioning in AD, and its relationship with memory performance.