Financial statement fraud (FSF) is costly for investors and can damage the credibility of the audit profession. To prevent and detect fraud, it is helpful to know its causes. The binary choice models (e.g. logit and probit) commonly used in the extant literature, however, fail to account for undetected cases of fraud and thus present unreliable hypotheses tests. Using a sample of 118 companies accused of fraud by the Securities and Exchange Commission (SEC), we estimated a logit model that corrects the problems arising from undetected frauds in U.S. companies. To avoid multicollinearity problems, we extracted seven factors from 28 variables using the principal factors method. Our results indicate that only 1.43 percent of the instances of FSF were publicized by the SEC. Of the six significant variables included in the traditional, uncorrected logit model, three were found to be actually non-significant in the corrected model. The likelihood of FSF is 5.12 times higher when the firm’s auditor issues an adverse or qualified report.