Measurements of currents through individual receptor-gated ion channels using the patch clamp technique have shown that each open channel contributes a square pulse of current and has a constant conductance, and that transitions between open and closed states are rapid1. To increase the signal-to-noise ratio of patch clamp recordings of glutamate channels in locust muscle, the bandwidth of the patch clamp amplifier has usually been restricted by a low pass filter (3 dB point, nominally 1 kHz with an 18 dB per octave slope). With concentrations of L-glutamate ≤5 × 10-5 M, we previously found that frequency histograms of channel lifetimes, constructed from data recorded with a 1 kHz bandwidth, could be fitted by single exponentials using the least-squares method2,3. Mean channel lifetimes estimated from the exponentials agree reasonably well with values obtained by 'noise' analysis of locust neuromuscular junctions4 and with mean values calculated directly by averaging the open times of single channel events5. However, the histograms are deficient in the region 0-1 ms because of the limited high-frequency response of the recording system. We have now recorded glutamate-gated channels in locust muscle membrane with signal-to-noise ratios of 3:1 or better and a recording bandwidth of d.c. to 10 kHz, and report lifetimes as brief as 50 μs. Frequency distribution histograms constructed from data recorded at 10 kHz suggest that the lifetime of the glutamate channel in extrajunctional membrane of locust muscle is governed by at least two exponential time-dependent processes, one of which has the effect of delaying channel closure.