Three uninterrupted, long (lasting respectively 7, 10, and 10 days) ASCA observations of the wellstudied TeV-bright blazars Mrk 421, Mrk 501, and PKS 2155[304 all show continuous strong X-ray Ñaring. Despite the relatively faint intensity states in two of the three sources, there was no identiÐable quiescent period in any of the observations. Structure function analysis shows that all blazars have a characteristic timescale of D1 day, comparable to the recurrence time and to the timescale of the stronger Ñares. On the other hand, examination of these Ñares in more detail reveals that each of the strong Ñares is not a smooth increase and decrease but exhibits substructures of shorter Ñares having timescales of D10 ks. We verify via simulations that in order to explain the observed structure function, these shorter Ñares ("" shots ÏÏ) are unlikely to be fully random, but in some way are correlated with each other. The energy dependent cross-correlation analysis shows that interband lags are not universal in TeV blazars. This is important since in the past only positive detections of lags were reported. In this work, we determine that the sign of a lag may di †er from Ñare to Ñare ; signiÐcant lags of both signs were detected from several Ñares, while no signiÐcant lag was detected from others. However, we also argue that the nature of the underlying component can a †ect these values. The facts that all Ñares are nearly symmetric and that fast variability shorter than the characteristic timescale is strongly suppressed, support the scenario where the light crossing time dominates the variability timescales of the day-scale Ñares.