The R20 neurons of Aplysia exhibit frequency-dependent spike broadening. Previously, we had used two-electrode voltage clamp to examine the mechanisms of this spike broadening (Ma and Koester, 1995). We identified three K+ currents that mediate action-potential repolarization: a transient A-type K+ current (I(Adepol)), a delayed rectifier current (IK-V), and a Ca(2+)-sensitive K+ current(IK-CA). A major constraint in that study was the lack of completely selective blockers for I(Adepol) and I(K-V), resulting in an inability to assess directly the effects of their activation and inactivation on spike broadening. In the present study, the dynamic-clamp technique, which employs computer simulation to inject biologically realistic currents into a cell under current-clamp conditions (Sharp et al., 1993a,b), was used either to block I(Adepol) or I(K-V) or to modify their inactivation properties. The data in this paper, together with earlier results, lead to the following hypothesis for the mechanism of spike broadening in the R20 cells. As the spike train progresses, the primary responsibility for spike repolarization gradually shifts from I(Adepol) to I(K-V) to I(K-Ca). This sequence can be explained on the basis of the relative rates of activation and inactivation of each current with respect to the constantly changing spike durations, the cumulative inactivation of I(Adepol) and I(K-V), and the progressive potentiation of I(K-Ca). Positive feedback interactions between spike broadening and inactivation contribute to the cumulative inactivation of both I(Adepol) and I(K-V). The data also illustrate that when two or more currents have similar driving forces and partially overlapping activation characteristics, selectively blocking one current under current-clamp conditions can lead to a significant underestimate of its normal physiological importance.