Jensen and Martin (1987) reported that subjects require additional time, and make more errors, when responding to a negative probe in a visual probe recognition task when the memory set contains one word that is either a rhyme of, differs by one letter from, or is a synonym of the probe. The experiments reported in this thesis investigate some aspect of each of these interference effects. Experiment 1 demonstrates that the phonological interference effect is not reduced when subjects must suppress articulation during a 12 second retention interval, and is not increased when subjects must recall the memory set after responding to the probe. This result implies that the phonological interference effect does not result from the retention of the memory set in an articulatory based short-term store. Experiment 2 and 3 demonstrate that similarity of graphemes, not just letters, leads to the graphemic interference effect since the effect was observed even when the probe and memory set were presented in different cases. Secondly, Experiments 2 and 3 demonstrate the suitability of a phonological control against which graphemic interference must be measured. Finally, Experiment 4 reveals that the semantic interference effect may result from directional association as well as synonymy. The results of these four experiments are consistent with several recent models of working memory proposed by Barnard (1985) and Monsell (1984) which propose that working memory involves both activated pre-existing memory traces as well as the temporary storage of trace images.