Under current process models of reasoning, detecting conflict between beliefs and logic is a key step that determines whether people will engage in reflective thinking. Conflict detection has been found across many tasks, but it is less reliably observed with syllogistic reasoning. ‘Reverse’ detection effects have been found in some studies, which are not easily explained within the current model of conflict detection and cannot be attributed only to measurement noise. In this study, we test whether ‘reverse’ detection effects in invalid syllogisms can be attributed to a mismatch between where conflict is thought to occur based on normative expectations and where conflict actually occurs for participants. We present evidence from two experiments (total N = 248) showing that invalid AC and DA syllogisms are intuitively valid to many participants, and that there are no differences in reaction time and confidence between valid and invalid items once believability and chosen response are taken into consideration. Further, we show that ‘reverse’ detection effects on invalid items disappear if we treat them as valid, on the assumption that this reflects where conflict occurs subjectively. These results indicate that conflict in these items likely occurs between two heuristic responses, rather than logic and heuristics; and that subjective, not objective conflict should be considered when measuring conflict detection.