Community colleges are typically assumed to be nonselective, open-access institutions. Yet access to college-level courses at such institutions is far from guaranteed: the vast majority of two-year institutions administer high-stakes exams to entering students that determine their placement into either college-level or remedial education. Despite the stakes involved, there has been relatively little research investigating whether such exams are valid for their intended purpose, or whether other measures of preparedness might be equally or even more effective. This paper contributes to the literature by analyzing the predictive validity of one of the most commonly used assessments, using data on over 42,000 first-time entrants to a large, urban community college system. Using both traditional correlation coefficients as well as more useful decision-theoretic measures of placement accuracy and error rates, I find that placement exams are more predictive of success in math than in English, and more predictive of who is likely to do well in college-level coursework than of who is likely to fail. Utilizing multiple measures to make placement decisions could reduce severe misplacements by about 15 percent without changing the remediation rate, or could reduce the remediation rate by 8 to 12 percentage points while maintaining or increasing success rates in college-level courses. Implications and limitations are discussed. (Contains 6 tables, 5 figures and 12 footnotes.)