We use detailed observational data constructed from daily passenger-level logbooks and weekly surveys to study the intertemporal labor supply decisions of Kenyan bicycle taxi drivers, while generating variation in cash on hand through randomized cash payouts. We document three key facts: (1) drivers work more in response to both unexpected and expected cash needs; (2) drivers discontinuously increase the probability of quitting once they have reached their day’s cash need; but (3) randomized cash payouts have no effect on labor supply. These results are consistent with models in which workers have reference-dependent preferences over earned income targets. A calibration exercise suggests that workers with such preferences earn about 5% more than they would with neoclassical preferences. We propose a model and interpretation of earned income targeting as morphine: it partially numbs the effort cost until the target is reached.