This paper examines the effects of performance pay on earnings using linked employee-employer panel data from Finland. These payroll data contain information on the exact share of earnings obtained and hours worked on a performance pay contract. Using these data, we estimate the effects of performance pay in the presence of both individual and firm-specific unobserved heterogeneity. Furthermore, we are able to estimate the effects of performance pay contracts in tasks of different complexity and for the subsample of workers who change jobs following an establishment closure. Unobservable firm characteristics explain 30-50% of the variance in performance pay. After controlling for unobservable individual and firm characteristics, performance pay workers earn substantially more than fixed rate workers. The effects persist when only workers who changed firms, and contracts, due to an establishment closure are used for identification. There is also a strong, negative relationship between job complexity and the incentive effects of performance pay. Finally, we exploit several 'natural experiments' where there was a compensation regime change in one plant of a given firm, but not in other plants. The plants are highly similar pre-regime change, and had a common trend in earnings pre-regime change. These experiments also yield substantial earnings premiums.