This study investigated the relationship between relational integration and semantic processing in numerical inductive reasoning using event-related potentials (ERPs). Participants were required to identify hidden rules in three-digit sequences and judged whether fourth probes were congruent with these rules. The experiment manipulated relational complexity: simple and hierarchical rules (e.g., ”12, 13, 14, 15” or ”12, 13, 15, 18”) shared identical first two numbers but differed in the third, in order to isolate the relational integration process. The type of cue was also manipulated: the no-cue condition presented only a fixation point, providing no rule information; the part-cue condition introduced operands (such as “+” or “–”) to ensure rules could be identified after presenting second numbers and confirmed after the onset of third numbers. The mass univariate ERP results locked to the third numbers revealed independent main effects of relational complexity and cue type on both the Late Negative Component (LNC) and Late Positive Component (LPC). The independent effects especially on the LNC implies that high-order numerical reasoning distinct from semantic processing.