We conducted a thorough investigation of the impact of representativeness on reasoning and metacognitive processes by employing the Linda problem. In congruent versions, the more representative response choice was also the correct one which was not the case in conflict versions. We manipulated the level of representativeness of the responses by making the component statements more or less representative of presented descriptions. The main findings revealed that higher levels of representativeness led to lower accuracy, faster responses and higher confidence. Detailed analysis revealed that certain reasoning processes heavily depend on representativeness and metacognitive assessments based on it (e.g., decoupling from a representative but incorrect response need not lead to lower confidence though it may be computationally expensive). Finally, responses remained mostly stable after rethinking, while answer changes towards the more representative choice were more likely than the opposite which implies Type 2 processes may support both normative and heuristic thinking.