Currents were obtained from single sodium channels in outside-out excised patches of membrane from the cell line GH3. The currents were examined in control patches and in patches treated with N- bromoacetamide ( NBA ) to remove inactivation. The single-channel current-voltage relationship was linear over the range -60 to + 10 mV, and was unaffected by NBA . The slope conductance at 9.3 degrees C was 12 pS, and the Q10 for single channel currents was about 1.35. The currents in both control and NBA -treated patches showed evidence of a slow process similar to desensitization in acetylcholine-receptor channels. This process was especially apparent at rapid rates of stimulation (5 Hz), where openings occurred in clusters of records. The clustering of records with and without openings was analyzed by runs analysis, which showed a statistically significant trend toward nonrandom ordering in the responses of channels to voltage pulses. NBA made this nonrandom pattern more apparent. The probability that an individual channel was "hibernating" during an activating depolarization was estimated by a maximum likelihood method. The lifetime of the open state was also estimated by a maximum likelihood method, and was examined as a function of voltage. In control patches the open time was mildly voltage-dependent, showing a maximum at about -50 mV. In NBA -treated patches the open time was greater than in the control case and increased monotonically with depolarization; it asymptotically approached that of the control patches at hyperpolarized potentials. By comparing channel open times in control and NBA -treated patches, we determined beta A and beta I, the rate constants for closing activation gates and fast inactivation gates. Beta I was an exponential function of voltage, increasing e-fold for 34 mV. beta A had the opposite voltage dependence. The probability of an open channel closing its fast inactivation gate, rather than its activation gate, increased linearly with depolarization from -60 to -10 mV. These results indicate that inactivation is inherently voltage dependent.