This paper investigates whether offers of commitment contracts, in the form of self-imposed choice-set restrictions and penalties with no financial upside, are well-targeted tools for addressing self-control problems. In an experiment on gym attendance (N= 1,248), we examine take-up of commitment contracts, and also introduce a separate elicitation task to identify actual and perceived time inconsistency. There is high take-up of commitment contracts for greater gym attendance, resulting in significant increases in exercise. However, this is take-up is influenced both by noisy valuation and incorrect beliefs about one’s time inconsistency. Approximately half of the people who take up commitment contracts for higher gym attendance also take up commitment contracts for lower gym attendance. There is little association between commitment contract take-up and reduced-form and structural estimates of actual or perceived time inconsistency. A novel information treatment providing an exogenous shock to awareness of time inconsistency reduces demand for commitment contracts. Structural estimates of a model of quasi-hyperbolic discounting and gym attendance imply that offering our commitment contracts lowers consumer surplus, and is less socially efficient than utilizing linear exercise subsidies that achieve the same average change in behavior.