EEG measures, especially alpha ERD and P3 ERP, has often been used as a tool for the investigation of the structure of categorical and conceptual representation of incoming stimuli. Studies of categorization that use alpha ERD and P3 signals typically have three characteristics: they use the oddball paradigm, the categories consist of discretely separable exemplars, and participants either have to categorize familiar objects or an intensive training precedes the test when EEG is recorded. We investigated whether alpha ERD and P3 are reliable indicators of category formation when instead of the structure of the already stable representation, the dynamics of learning of novel categories is investigated, and if so, whether they indicate the same underlying cognitive processes. To this end, we altered the usual oddball categorization paradigm in two aspects. First, participants had to categorize schematic human silhouettes ranging from slender to chunky in two categories (Thin and Fat), that is stimuli varied continuously along the parameter of width. Specifically, category exemplars were distributed as two Gaussians per category, one closer to the category boundary, the other at the extreme of the range. Second, we recorded neural responses with high-density (128-electrode) nets from the very beginning of the category formation. Apart from EEG recordings, we collected implicit and explicit behavioral responses (RT, mean and subjective uncertainty reports) to assess the correspondence between the two. Confirming earlier results, we found significant differences in alpha ERD elicited by frequent and infrequent stimuli [t(11)=2.66, p<.02]. Importantly, learning was reflected in the difference of alpha-band suppression between neural responses collected early vs. late during the experiment, especially for exemplars at the extremes of stimulus range [t(11)=2.98, p = .012]. This difference was less articulate for stimuli sampled around the category boundary where confidence was also lower, suggesting that alpha ERD and subjective confidence in acquired knowledge are correlated. We found no significant differences in P3 amplitudes neither at the boundary nor at the extremes. Thus, alpha-suppression is not only a reliable measure of the already acquired structure of concepts, but it also allows for tracking the ongoing acquisition of categories. In contrast, P3, seems to be a different measure of category representations, and it is highly sensitive to the discrete—continuous distinction of the stimuli.