Total sleep deprivation (TSD) causes a decline in almost all cognitive domains, especially working memory. However, we do not have a clear understanding of the degree working memory is impaired under prolonged TSD, nor do we know the underlying neurophysiological mechanism. In this study, we recorded EEG data from 64 subjects while they performed a working memory task during resting wakefulness, after 24 h TSD, and after 30 h TSD. ANOVA was used to verify performance differences between 24 h and 30 h TSD in working memory tasks: (1) reaction time and accuracy hit rates, (2) P200, N200, and P300 amplitude and latency in measurements of event-related potential, as well as (3) effective connectivity strength between brain areas associated with working memory. Compared to 24 h TSD, 30 h TSD significantly decreased accuracy hit rates and induced a larger N200 difference waveform. The effective connectivity analysis showed that 30 h TSD also decreased beta frequency in effective connection strength from the right insular lobe to the left anterior cingulate cortex (ACC). Effective connection from the left ventrolateral prefrontal cortex to the left dorsolateral prefrontal cortex increased in the match condition of the 2-back task. In conclusion, 30 h TSD had a greater negative impact on working memory than 24 h TSD. This impairment of working memory is associated with decreased strength in the effective connection from the right insula to the left ACC.