Potassium currents were recorded from voltage-clamped nodes of isolated, myelinated axons of Rana pipiens. Nodes were maintained in a modified Ringer solution containing tetrodotoxin to block sodium current and 47.5 mM-potassium to minimize effects of extracellular potassium accumulation. Voltage protocols included depolarizing pulses lasting a few milliseconds to several seconds. Fluctuations about the ensemble average of the current were characterized in terms of non-stationary variance and autocovariance. The fluctuations had a Gaussian amplitude distribution and were virtually free of contaminations from systematic variations of the membrane current. Corrections for background noise were based on measurements done while potassium current was blocked with tetraethylammonium, and on simulations of extrinsic current fluctuations expected to arise from noise in the actual membrane voltage. The fluctuations were attributed to variations of nodal potassium conductance, since they were absent at the reversal potential of the potassium current and at membrane voltages that do not activate potassium current. Covariances indicated that voltage steps that reversed a macroscopic potassium current also reversed the sign of the fluctuation. Plots of the conductance variance versus the mean potassium conductance were generated from both the activation and deactivation (tail) phases of the potassium currents at various voltages between -80 and +70 mV. When the current was activated by a small depolarization (-50 mV) the trajectories from both phases were indistinguishable and were fitted by the parabola expected for a single population of channels with only one open-channel conductance. Apparent single-channel conductance from the early activation phase averaged 24 pS and was not significantly voltage dependent. In contrast, experiments with large depolarizations (+10 to +70 mV) gave significantly different variance--mean trajectories during activation and deactivation and these trajectories were poorly fitted by parabola. This result indicates that the fluctuations reflect several populations of channels and/or a population of channels that can have several levels of non-zero conductance. Projections of the fluctuation covariance showed long correlations, as well as the rapidly decaying component expected from the activation gating of channels. A slow fluctuation arose at a time slightly later than the rise of potassium current, spanned the entire length of brief depolarizations, and extended up to 880 ms during long depolarizations.(ABSTRACT TRUNCATED AT 400 WORDS)