This paper considers identification and estimation of the effect of a mismeasured binary regressor in a nonparametric or semiparametric regression, or the conditional average effect of a binary treatment or policy on some outcome where treatment may be misclassified. Failure to account for misclassification is shown to result in attenua- tion bias in the estimated treatment effect. An identifying assumption that overcomes this bias is the existence of an instrument for the binary regressor that is conditionally independent of the treatment effect. A discrete instrument suffices for nonparametric identification. THIS PAPER PROVIDES CONDITIONS for identification and estimation of the av- erage effect, conditioned on covariates, of a binary treatment, program, or pol- icy on a scalar outcome when treatment may be misclassified. More generally, what is provided is an estimator of the effect of a binary regressor in a con- ditional mean regression (which may also include other regressors) when the binary regressor is observed with error. This equals the conditional average treatment effect given a weak unconfoundedness assumption. Misclassification occurs when a binary variable (the treatment indicator) is measured with error, that is, some units are reported to have received treat- ment when they actually have not, and vice versa. For example, in a returns to schooling analysis the outcome could be wages, the binary variable could be attaining a college degree, and misclassification could arise from school tran- script reporting errors. See, e.g., Kane and Rouse (1995) and Kane, Rouse, and Staiger (1999). Bollinger (1996) considers misclassification of union status in wage equations. If treatment is a job training program, misclassification may arise if individuals assigned to the program fail to attend or if those not as- signed obtain training elsewhere. Similarly, for medical treatment, individuals assigned a drug might fail to take it and those assigned a placebo might obtain treatment from another source. More generally, misclassification describes any binary variable that is sometimes mismeasured. This paper first shows that misclassification causes attenuation bias in esti- mated treatment effects, analogous to the attenuation bias of classically mis- measured variables in linear regression models. Additional assumptions on