The research originated from a noise quality problem common with Diesel powered cars, where impulsive, repetitive combustion noise is perceived as particularly unpleasant by passengers and pedestrians. The main characteristic of combustion noise is that it consists of short duration pulses, and it was desirable to understand how these short duration pulses could be masked, e.g. by background noise. This research therefore addresses the question, whether the critical band/auditory filter mechanism remains functional when the duration of the probe tone is decreased. Frequency selectivity is measured using Patterson's notched noise method using three probe tone durations (400 ms, 40 ms and 4 ms). Five psychoacoustic threshold experiments are carried out with 3, 20, 4, 1 and 10 subjects respectively (38 subjects in total). The listeners had to detect a 2-kHz probe tone in a notched noise masker at 30 dB/Hz spectrum level, centred on the tone. Thresholds are measured with the method of adjustment, where the subject is asked to adjust the level of the probe tone to masked threshold. Stimuli are mainly presented via loudspeakers in an anechoic chamber to both ears, but also monaurally and binaurally over headphones. All notched noises are synthesized digitally on a computer by adding up sine waves with random phase. The resulting threshold-versus-notch-width-curves are plotted and compared for all three probe tone durations. The steepness of these curves is taken as a measure of frequency selectivity (auditory filter width). It was found that the curves are very similar for all three durations, indicating that the frequency selective mechanism is maintained for signal durations down to 4 ms and 1 ms.