Governments are charged with monitoring citizens' compliance with prescribed behavioral standards and punishing noncompliance. Flaws in information available to enforcing agents, however, may lead to subsequent enforcement errors, eroding government authority and undermining incentives for compliance. We explore these concepts in a laboratory experiment. A "monitor" makes punishment decisions after receiving noisy signals about individuals' choices to contribute to a public good. We find that the possibility of wrongly accusatory signals has a more deleterious effect on contribution levels than the possibility of wrongly exculpatory signals, contrary to a rational-choice benchmark model predicting equal effects. We trace this across-treatment difference to a "false positives trap": when members of a largely compliant population are sometimes incorrectly accused, some will be unjustly punished if enforcement power is employed, but non-compliant individuals will escape punishment if that power is abdicated. Either kind of error discourages compliance. An additional treatment demonstrates that the effectiveness of a given enforcement institution may vary, depending on its origins. We consider implications of our findings for theories of deterrence, fairness, and institutional legitimacy.