The Peter Principle states that, after a promotion, the observed output of promoted employees tends to fall. Lazear (2004) models this principle as resulting from a regression to the mean of the transitory component of ability. Our experiment reproduces this model in the laboratory by means of various treatments in which we alter the variance of the transitory ability. We also compare the efficiency of an exogenous promotion standard with a treatment where subjects self-select their task. Our evidence confirms the Peter Principle when the variance of the transitory ability is large. In most cases, the efficiency of job allocation is higher when using a promotion rule than when employees are allowed to self-select their task. This is likely due to subjects’ bias regarding their transitory ability. Naïve thinking, more than optimism/pessimism bias, may explain why subjects do not distort their effort prior to promotion, contrary to Lazear’s (2004) prediction.