Transient global amnesia (TGA) is usually considered to produce a profound impairment of long-term episodic memory, while at the same time sparing working memory. However, this neuropsychological dissociation has rarely been examined in detail. While a few studies have assessed some components of working memory in TGA, the results that have been obtained are far from conclusive. To clarify this issue, we carried out a comprehensive investigation of working memory in 10 patients during a TGA attack. In the first study, we report the results from three patients examined with a battery of neuropsychological tests designed to assess each of the three subcomponents of Baddeley's model of working memory. In a second study, seven different patients underwent neuropsychological investigations that focused specifically on the central executive system, using a protocol derived from a study by Miyake and colleagues. Our findings showed that subcomponents of working memory, such as the phonological loop and visuo-spatial sketch pad, were spared in TGA patients. Specific executive functions that entailed inhibitory control, dual task performance, updating and shifting mechanisms were also found to be normal. However, we found significantly impaired performance for the Brown-Peterson test, and that TGA patients were significantly impaired in the recollection of their episodic memories. They also made reduced numbers of 'remember' compared with 'know' judgments in the episodic memory test several days after TGA. On the basis of our findings, it would appear that the episodic memory deficit during TGA is not related to elementary aspects of executive functioning. Our data also highlight the nature of the cognitive mechanisms involved in the Brown-Peterson task, which may well depend on long-term memory (such as the process of semantic encoding). Lastly, the selective deficit in recollective episodic memories observed in TGA may be principally related to medial temporal lobe abnormalities that have been reported in this syndrome.