|We use the American Time Use Survey (ATUS) 2003-2012 to estimate time spent in non-work on the job. Non-work is substantial and varies positively with local unemployment. Time spent in non-work conditional on any positive amount rises, while the fraction of workers reporting positive values declines with unemployment. Both effects are economically important, and are consistent with a model in which heterogeneous workers are paid efficiency wages. That model correctly predicts the relationship between the incidence of non-work and unemployment benefits in state data linked to the ATUS, and is consistent with estimated occupational differences in non-work incidence and intensity, as well as the cyclical behavior of aggregate labor productivity.