1. The time constants of motor and sensory nerve fibres were studied in normal human ulnar nerves by the method of latent addition, using threshold tracking to follow the recovery of excitability after brief conditioning current pulses. The 60 microseconds test and conditioning stimuli were applied at the wrist, and the conditioning stimuli were set to 90, 60, 30, ‐30, ‐60 and ‐90% of the control threshold current. Compound muscle action potentials were recorded from abductor digiti minimi, and sensory nerve action potentials from the little finger. 2. Recovery from depolarizing conditioning pulses was slower than recovery from hyperpolarizing pulses and strongly dependent on conditioning pulse amplitude. The voltage dependence of latent addition was attributed to subthreshold activation of sodium channels (local response). 3. Motor and sensory nerve excitability generally recovered from ‐90% hyperpolarizing pulses as the sum of two exponential components, although the slow component was negligible in some motor nerves. The fast component (time constant 43.3 +/‐ 2.0 microseconds, mean +/‐ S.E.M., n = 9) was similar between motor and sensory fibres in the same subject. It showed no consistent voltage dependence, and was attributed to a passive input time constant of the fibres. The slow component of recovery from hyperpolarizing pulses was greater in sensory than in motor fibres and was voltage dependent: it could be greatly increased in motor and sensory fibres by steady depolarization. It was attributed to a regenerative membrane current, active at the resting potential in sensory and at least some motor nerves. 4. The latent addition responses were compared with the computed responses of four theoretical models. Both motor and sensory responses were well fitted by a model in which a fraction of the sodium channels (less in motor than in sensory fibres) were activated at potentials 20 mV more negative than normal and at half the normal rate, and did not inactivate. 5. It is concluded that the differences in latent addition between motor and sensory fibres are primarily due to differences in non‐classical, voltage‐dependent ion channels, active close to the resting potential. These "threshold channels' may help to account for the longer strength‐duration time constant of sensory fibres, for their lower rheobase, and for their greater tendency to fire repetitively.