Buffer versus embedded processes accounts of short-term memory (STM) have been under debate for decades. Buffer models propose dedicated storage for maintaining different types of information, where these buffers are separate from long-term memory (LTM) for those kinds of information. In contrast, embedded processes models argue against the existence of buffers, and claim STM consists of the activated portion of LTM. This thesis tested two competing models for verbal STM and obtained evidence from three experiments with novel multivariate neuroimaging and non-invasive brain stimulation approaches. From the perspective of cognitive neuroscience, buffer models predict that storage buffers depend on brain regions different from the LTM regions used to represent permanent knowledge, whereas embedded processes models predict that STM recruits the same neural substrate as LTM. Experiment 1 used functional magnetic resonance imaging with representational similarity analysis (RSA) to examine the correspondence of multi-voxel neural activation patterns with the theoretical representations for both phonological and semantic STM. In the phonological domain, a speech processing region in the left superior temporal gyrus (STG) showed RSA evidence of phonological coding during the encoding period, but not during the delay period. In contrast, the left supramarginal gyrus (SMG) showed RSA evidence of phonological storage during the delay period. In the semantic domain, the triangular part of the left inferior frontal gyrus showed RSA evidence of semantic coding during the encoding period, but not during the delay period. In contrast, some posterior regions such as the left angular gyrus, the left posterior middle temporal gyrus and an anterior region in the left middle frontal gyrus showed RSA evidence of semantic retention during the delay period, with the angular gyrus allowing for decoding of either phonological and semantic STM, depending on the task context. Results of Experiment 1 illustrated that different regions were involved in encoding and maintenance for phonological and semantic STM respectively. Experiment 2, using nonword stimuli, tested distractor interference effects on phonological STM. Although the speech processing region in the left STG showed RSA evidence of phonological storage for these materials during the delay period, such evidence was absent when distractors were presented. However, the proposed buffer region in the left SMG showed RSA evidence of phonological retention even in the presence of distractors during the delay period. Results of Experiment 2 suggested that the buffer region played a more important role in phonological STM in the face of distracting stimuli. Experiment 3 used transcranial magnetic stimulation (TMS) to test the necessity of the brain regions implicated in phonological STM from Experiments 1 and 2. An effect of TMS on response time for a STM recognition task was observed when the left SMG was stimulated during the delay period, whereas stimulation at the left STG or an occipital control region had no effect on behavioral performance. Results of Experiment 3 confirmed the causal role of the left SMG in phonological STM. Taken together, converging evidence from three experiments provided greater support for a buffer account than an embedded processes account of verbal STM. General implications for the buffer vs. embedded processes debate, as well as implications for theories of the neural basis of working memory are discussed.