The ability to form judgments about one's own cognitive abilities, known as metacognition, has been extensively studied recently. Metacognitive judgments are not always accurate, but the extent to which individuals can evaluate the quality of their own metacognition remains unclear. To explore this ability, termed meta-metacognition, we present here two experiments. Observers were asked to judge which of two perceptual decisions was more likely to be correct (a metacognitive choice) and to evaluate their confidence in such choices (a meta-metacognitive judgment). By collecting confidence ratings after each perceptual decision and evaluating how individuals make probabilistic inference in a separate task, we demonstrated two key empirical findings. First, meta-metacognitive judgments predicted the quality of metacognition beyond the information that can be optimally extracted from perceptual confidence. Second, participants' metacognitive choices contradicted their ratings of perceptual confidence in nearly 25% of cases, and when this occurred, their metacognitive accuracy mostly improved.