We show the extent of errors made in the award of disability insurance using matched survey-administrative data. False rejections (Type I errors) are widespread, and there are large gender differences in these type I error rates. Women with a severe, work-limiting, permanent impairment are 20 percentage points more likely to be rejected than men, controlling for the type of health condition, occupation, and a host of demographic characteristics. We investigate whether these gender differences in Type I errors are due to women being in better health than men, to women having lower pain thresholds, or to women applying more readily for disability insurance. None of these explanations are consistent with the data. We use evidence from disability vignettes to suggest that there are different acceptance thresholds for men and women. The differences by gender arise because women are more likely to be assessed as being able to find other work than observationally equivalent men. Despite this, after rejection, women with a self-reported work limitation do not return to work, compared to rejected women without a work limitation.