Processing and maintenance in working memory involve active attention allocation; consequently, it is possible that a recognition process could interfere with the performance of highly demanding working memory tasks. Event-related brain potentials (ERPs) were recorded while fourteen healthy male adults performed a visual verbal dual working memory task. Three conditions were examined: A) reference (with no preceding stimuli); B) happy, angry or neutral faces presented 250 ms prior to task onset for 30 ms; and, C) visual noise control condition. Behavioral results showed that reaction times were significantly longer in the condition preceded by the presentation of faces, regardless of their emotional content. ERPs showed a predominantly right temporo-occipital negative component at around 170 ms during conditions B and C (higher amplitude in B), which probably reflects the pre-categorical structural encoding of the face. Succeeding task-onset, an early positive right temporo-parietal component (P380) appeared during condition B, probably reflecting a delayed reallocation of working memory attentional resources to carry out the task requirements. Afterwards, two other main fronto-parietal components were observed in the three conditions: a positive wave that peaked at around 410 ms, and a subsequent negative component (N585). This latter waveform reached a significantly higher amplitude during the reference condition (A) and was interpreted as mirroring the phonologic-lexical manipulation of the stimuli in working memory. These results suggest that early face detection could induce an attentional decrement that interfere a subsequent visual verbal working memory task performance. They also suggest that while face detection and facial emotional content analysis might be parallel processes, they are not onset-synchronized.