The revelation effect is a change in response behavior induced by a preceding problem-solving task. Previous studies have shown a revelation effect for faces when the problem-solving task includes attractiveness ratings of the faces. Immediately after this problem-solving task participants judged faces as more familiar than without the problem-solving task. We replicated this result in Experiment 1. Based on the discrepancy-attribution hypothesis, we predicted that a problem-solving task that excludes attractiveness ratings would not elicit a revelation effect. However, we found a reversed revelation effect with a problem-solving task that required participants to solve a puzzle of each face (Experiments 2-3). In Experiments 2 and 3, participants judged faces as "less" familiar after the puzzle task. Our findings support the notion that the revelation effect may manifest as either an increase or a decrease of the experienced familiarity towards the recognition probe. However, our results contradict all current theories of the revelation effect. We discuss implications of our findings for revelation effect theories and provide a possible explanation. (Contains 2 tables and 1 figure.)