Extant theory suggests that metacognitive judgments are informed by both privileged metacognitive cues and observable, external cues that can be evaluated via non-privileged mechanisms, such as statistical regularity monitoring and top-down, domain-specific beliefs. Across three studies implementing a variant of the learner-observer-judge paradigm, we explore the extent to which decision makers and observers rely on each of these mechanisms to infer the processes underlying a decision maker’s multi-attribute choice behavior. In Study 1, we demonstrate that decision makers (n = 221) more accurately estimate the weights they applied to each attribute in a multi-attribute choice task than observers (n = 220) who studied their choices. This contrast highlights the value of privileged metacognitive cues. However, observers’ estimates were still weakly correlated with decision makers’ true weights, suggesting that reasonably accurate inferences can be made, even without access to privileged metacognitive cues. In Study 2, we demonstrate that the relative (but not absolute) accuracy of observers’ (n = 218) inferences decreases when the context of the decision is obscured, suggesting that domain-specific beliefs slightly improve metacognitive inferences above and beyond what can be inferred via statistical regularity monitoring alone. In Study 3, we replicate these findings and investigate differences in subjective experiences (e.g., motivation, effort) across conditions. Together, our findings suggest that privileged metacognitive cues do improve the accuracy of decision makers’ metacognitive knowledge, but that observers can make reasonably accurate inferences about decision makers’ choice behavior, particularly when they can combine statistical regularity monitoring with domain-specific beliefs. .