Overtesting induces unnecessary yield loss of VLSIs. Untestable faults, which have no effect on the normal functions of circuits, may be detected in scan testing through scan chains. In this case, the detected untestable faults cause overtesting. Untestable faults consist of uncontrollable faults, unobservable faults, and uncontrollable and unobservable faults. Uncontrollable faults may be detected under invalid states through scan chains by shift-in operations. Unobservable faults cannot be observed at primary outputs, but their effects may be propagated to scan flip-flops. Thus, unobservable faults may be detected through scan chains by shift-out operations. Several methods to reduce the number of detected untestable faults have been proposed. These methods identify invalid states and generate test patterns to avoid invalid states. As a result, the number of detected uncontrollable faults are reduced. However, these methods cannot reduce the number of detected unobservable faults. In this paper, both uncontrollable and unobservable faults are identified using a multi-cycle capture test generation method. We evaluate the relationship between the numbers of untestable faults and the number of time expansions for ISCAS'89 benchmark circuits, and also evaluate factors to identify and classify untestable faults.