During observation of an ambiguous figure perception becomes unstable and alternates between different interpretations. Tiny low-level changes can disambiguate an ambiguous figure and thus stabilize its percept. We compared ERPs evoked by ambiguous stimuli and disambiguated stimulus variants across different visual categories (geometry, motion) and complexity levels (up to emotional face expressions). Disambiguated stimulus variants cause stable percepts and evoke much larger amplitudes of two positive ERP components than ambiguous stimuli (d > 1). This pattern of results is highly consistent across visual categories and complexity levels. The generality of our findings across categories and complexity levels points to higher-level/cognitive mechanisms: Given a priori incomplete, noisy and ambiguous sensory input, we postulate a high-level Bayesian inference unit that evaluates the reliability of perceptual processing results. Large ERP amplitudes would reflect high perceptual reliability.