While earnings risk is essentially subjective, it is typically inferred from administrative data. We introduce a survey to measure subjective earnings risk, paying particular attention to the expected impacts of job transitions on earnings. Linking with administrative data provides multiple credibility checks. Subjective expectations about earnings growth and job transitions are consistent with actual realizations when appropriately aggregated. We also find subjective earnings risk is lower than risk inferred from administrative data because expected earnings growth is heterogeneous, even within narrow population groups. A life-cycle search model calibrated to the administrative data can recover the basic patterns of subjective risk.