Humans carry out a variety of complex cognitive processes, such as language comprehension, learning, and reasoning. These cognitive functions all require temporal maintenance and manipulating information that is no longer presented. Our research interest focuses on the ability to temporarily hold the information in mind, called working memory (WM) maintenance. Since WM maintenance is likely to be essential for daily activities, it is crucial to understand the underlying mechanisms. Notwithstanding extensive behavioral and neural studies for almost a century, the exact mechanisms underlying WM remain unclear. However, the new machine learning (ML) technique known as “Multivariate Pattern Analysis (MVPA)” seems to offer valuable insights into where information is stored in the brain during WM maintenance. Moreover, MVPA on electroencephalographic (EEG) contributed to cognitive research by decoding brain activity patterns across various conditions and materials. This approach indicated significant potential for decoding WM content, but some recent (unpublished) studies reported cases of unsuccessful decoding of WM content. The goal of the study was to “decode” (1) the nature (category) of information being maintained in WM and (2) the strategy used to maintain information in WM using neuroimaging and ML techniques. To address these goals, we initially conducted a systematic literature review to identify a successful MVPA pipeline to apply to EEG data during behavior tasks. Next, the pipeline applied EEG data recorded during the visual delay-recognition task to decode three different visual categories (Visual, Spatial, and Verbal) shown in the experiment. As a result, we successfully decoded the category of information held in WM during the maintenance period. Furthermore, we tested whether the same MVPA pipeline could also achieve above-chance decoding of the four strategies (Refresh, Rehearsal, Imagery, and Sentence Elaboration) instructed to maintain words in WM. Contrary to our expectations, we could not decode strategies used during the WM maintenance period, but rehearsal conditions did seem to result in a higher accuracy rate than the other strategies. Our findings suggest that WM content can be successfully decoded using EEG MVPA techniques. Further, the different WM maintenance strategies do not seem to show significant distinctions in MVPA, but rehearsal might operate differently from the others. Our results indicate that MVPA can be a potential tool to investigate the WM maintenance representations, but future studies are needed to understand the impact of pipeline decisions on successful decoding.