While the video gaming training may enhance visual working memory (VWM), at the same time VWM is a key cognitive function in effective video gaming. In our study we examined the relationship between EEG data obtained during the visual working memory task and a real-time strategy (RTS) video game performance. The training lasted 30 hour, during which participants played StarCraft II, in either a fixed (FEG) or variable environment (VEG) training model, which were made to explore the role of a training complexity as an important factor. An EEG measurement took place before and after the training while performing a visual working memory task. Initial (pre-training) posterior alpha and frontal midline theta power have been specified as predictors of players' in-game advancement. Using a logistic regression model we determined telemetric variables predictive of game output. It turned out that both oscillatory bands were predictive of values of the four of the previously determined telemetric variables, but only in the FEG. Moreover we have seen differences between FEG and VEG in telemetry as well as in the neurophysiological data. Our results show how important is the complexity of the training regimen for observing the predictive power of VWM’s EEG oscillatory indicators.