We show how using censored regressors leads to expansion bias, or estimated effects that are proportionally too large. We show the necessity of this effect in bivariate regression and illustrate the bias using results for normal regressors. We study the bias when there is a censored regressor among many regressors, and we note how censoring can work to undo errors-in-variables bias. We discuss several approaches to correcting expansion bias. We illustrate the concepts by considering how censored regressors can arise in the analysis of wealth effects on consumption, and on peer effects in productivity.