Many policy design issues depend crucially on the nature of the idiosyncratic risks to labour income. The earning dynamics literature has typically relied on an implicit or explicit assumption that earnings shocks are log-normally distributed. This column challenges conventional knowledge by bringing in new evidence from a very large administrative dataset on US workers. It presents evidence suggesting income shocks exhibit substantial deviations from log-normality, and that shock persistence depends on income levels as well as the size and sign of the shock.