Around 65% in the classic (most-cited) version of Stanley Milgram’s obedience experiments; lower (roughly 40% average) across all his variations.

In Milgram’s best-known 1960s studies at Yale, ordinary participants were instructed by an authority figure (an experimenter in a lab coat) to administer what they believed were increasingly severe electric shocks to a learner for wrong answers. The learner was an actor who protested, complained of a heart condition, and eventually went silent. In the baseline condition:

• 65% (26 out of 40 participants) continued all the way to the maximum 450 volts.
• Every participant went at least to 300 volts.

This was far higher than experts (including psychiatrists) had predicted beforehand. Similar rates appeared in some replications decades later.

Important caveats

Milgram ran many variations (over 20 conditions involving hundreds of participants total). Obedience rates ranged widely:

• Up to ~90%+ in some setups (e.g., when the participant did not have to directly administer the shocks).
• Down to 0% in others (e.g., when two authorities gave conflicting orders, when the experimenter was the victim, or when peers openly refused).
• Average across conditions: roughly 37–44%.

Factors that strongly reduced obedience included greater proximity to the victim, the absence or reduced legitimacy of the authority figure, and the presence of dissenting peers.

The finding does not mean 65% of people will blindly obey any order in everyday life “regardless of content.” The experiments involved a specific laboratory context with a perceived legitimate scientific authority, gradual escalation, and diffusion of responsibility. Real-world obedience depends heavily on the situation, perceived legitimacy of the authority, personal values, social support for resistance, and many other variables.

In short: the famous headline number is ~65% fully obedient in the standard setup, but the broader body of results shows substantial variation and that disobedience was common under many conditions.