The time course of recovery from use-dependent block of sodium channels caused by local anesthetics was studied in squid axons. In the presence of lidocaine or its quaternary derivatives, QX-222 and QX-314, or 9-aminoacridine (9-AA), recovery from use-dependent block occurred in two phases: a fast phase and a slow phase. Only the fast phase was observed in the presence of benzocaine. The fast phase had a time constant of several milliseconds and resembled recovery from the fast Na inactivation in the absence of drug. Depending on the drug present, the magnitude of the time constant of the slow phase varied (for example at -80 mV): lidocaine, 270 ms; QX-222, 4.4 s; QX-314, 17 s; and 9-AA, 14 s. The two phases differed in the voltage dependence of recovery time constants. When the membrane was hyperpolarized, the recovery time constant for the fast phase was decreased, whereas that for the slow phase was increased for QX-compounds and 9-AA or unchanged for lidocaine. The fast phase is interpreted as representing the unblocked channels recovering from the fast Na inactivation, and the slow phase as representing the bound and blocked channels recovering from the use-dependent block accumulated by repetitive depolarizing pulse. The voltage dependence of time constants for the slow recovery is consistent with the m-gate trapping hypothesis. According to this hypothesis, the drug molecule is trapped by the activation gate (the m-gate) of the channel. The cationic form of drug molecule leaves the channel through the hydrophilic pathway, when the channel is open. However, lidocaine, after losing its proton, may leave the closed channel rapidly through the hydrophobic pathway.