Studies on human reasoning have long concluded that intuitions can bias inference and result in violations of logical principles. Prominent dual-process models, which portray thinking as an interplay between intuitive (Type 1) and deliberate (Type 2) thought processes, have offered a reasonable account for this observation. According to this explanation, logical reasoning is an example of a task that demands effortful deliberate thinking. However, in recent years, amassing evidence obtained with novel experimental paradigms have brought into question the traditional dual-process characterization. A crucial finding is that people can process logical principles in classic reasoning tasks intuitively and without deliberation. In accordance with this idea, an alternative hybrid model has been proposed (Pennycook, Fugelsang & Koehler, 2015). According to this modern view, which was also a theoretical framework for the present study, people's intuitive reasoning performance and experience of conflict are considered to be determined by the absolute and relative strength of two competing intuitions - heuristic and logical. The smaller the strength disparity between the two intuitions, the greater the conflict. Conflict detection is deemed to be an initiator of analytic engagement. Many researchers have recently taken interest in illuminating the nature of conflict monitoring and control. The framework that allows them to do so is called meta-reasoning (Ackerman i Thompson, 2014), and it is focused on explaining the processes that underlie the formation of metacognitive judgments. The present study aimed to examine the influence of instruction type (reasoning according to belief or statistics) on the performance and metacognitive judgments in the base rate tasks. The instruction type was varied to affect the strength of the logical intuition that is hypothesized to cue selection of the base-rate response. The data were obtained utilizing the two response paradigm. Participants were required to give an initial response under time-pressure, which allows for identifying the presumed intuitively generated question. The results show that participants in both groups are quicker to respond to congruent than to incongruent problems. Also, they are more confident in the accuracy of congruent than incongruent answers. The findings, which are generally in line with the proposed hybrid model, suggest that, regardless of the instruction type, conflict detection in incongruent tasks serves as a bottom-up source of analytic processes.