We have investigated slow inactivation in a rat axonal K+ channel, the I channel. Using voltage steps to potentials between -70 mV and +80 mV, from a holding potential of -100 mV, we observed a marked slowing of inactivation at positive potentials: the time constant was 4.5 +/- 0.4 s at -40 mV (mean +/- S.E.M.), increasing to 14.7 +/- 2.0 s at +40 mV. Slowed inactivation at positive potentials is not consistent with published descriptions of C-type inactivation, but can be explained by models in which inactivation is preferentially from closed states (which have been developed for Kv2.1 and some Ca2+ channels). We tested two predictions of preferential closed-state models: inactivation should be more rapid during a train of brief pulses than during a long pulse to the same potential, and the cumulative inactivation measured with paired pulses should be greater than the inactivation at the same time during a continuous pulse. The I channel does not behave according to these predictions, indicating that preferential closed-state inactivation does not explain the slowing of inactivation we observe at positive potentials. Inactivation of the I channel therefore differs both from C-type inactivation, as presently understood, and from the inactivation of Kv2.1.