<p>This study adopts the Think/No-Think (TNT) paradigm for the first time to examine whether activation of thematic semantic relation representations is intrusive. In Experiment 1, during the learning phase, participants memorized cue–target word pairs, with half being thematically related and the other half taxonomically related. In the Think/No-Think phase, both relation types were equally distributed across the Think, No-Think, and baseline conditions. In the Think condition, participants retrieved the target words, while in the No-Think condition, they suppressed the target words and reported the frequency of intrusions. The final phase included both a probe recall test and a target recognition test. Results showed that in the probe recall test, recall performance for thematic No-Think pairs did not differ from baseline, whereas a significant suppression effect was observed for taxonomic pairs. In addition, during the No-Think phase, intrusion frequency for thematic pairs (self-reported) was significantly higher, with no significant reduction in intrusions over time. Experiment 2 further examined the modulation of this intrusiveness by the strength of thematic semantic association. Results showed that in the cue recall test, unlike the strong-association thematic condition, the recall accuracy of No-Think word pairs in the weak-association thematic condition was significantly lower than that of the baseline pairs. In addition, during the No-Think phase, intrusion frequency (self-reported) for weak-association thematic pairs was significantly lower, with a significant reduction in intrusions over time. All findings indicate that activation of thematic semantic relationship representations is intrusive, but is modulated by the strength of thematic semantic association.</p>