We model that separation is a rational decision to resolve the inherent uncertainty about quality form the matching process. In health insurance markets, quality of services is revealed to consumers only after enrollment. Discrepancies between the expected and realizeed quality lead utility-maximizing enrollees to disenroll if they find a better alternative. Accordingly, factors that reduce this discrepancy will decrease disenrollment. The firm-level empirical analysis shows that disenrollment is relatively small in markets where the reputation effect works efficiently becuse consumers can predict the expected quality accurately.