Earnings risk is central to economic analysis. While this risk is essentially subjective, it is typically inferred from administrative data. Following the lead of Dominitz and Manski (1997), we introduce a survey instrument to measure subjective earnings risk. We pay particular attention to the expected impact of job transitions on earnings. A link with administrative data provides multiple credibility checks. It also shows subjective earnings risk to be far lower than its administratively- estimated counterpart. This divergence arises because expected earnings growth is heterogeneous, even within narrow demographic and earnings cells. We calibrate a life-cycle model of search and matching to administrative data and compare the model-implied expectations with our survey instrument. This calibration produces far higher estimates of individual earnings risk than do our subjective expectations, regardless of age, earnings, and whether or not workers switch jobs. This divergence highlights the need for survey-based measures of subjective earnings risk.