The behaviour of many animals indicates a high precision in the processing of sensory signals, which often seems at odds with the large variability of individual neuronal responses. Using the directional hearing of the grasshopper Chorthippus biguttulus (Ch. biguttulus) as a model system, we investigated the possible contributions of temporal integration and parallel processing at the receptor level to the observed behavioural acuity. The precision of the animals' phonotactic orientation behaviour to stimuli of different durations was measured and compared with the spiking responses of auditory receptor cells, leading to the following results. A behavioural decision based on integrating the spikes from a pair of receptors (one cell in each ear) over a 1000-ms time window would substantially decrease the error probability, compared with evaluating a 250-ms period. The animal as a whole, however, responded as precisely to a 250 ms stimulus as to a stimulus of natural duration (c. 1000 ms). A phonotactic decision based on a 1.5 dB intensity difference between the ears corresponds to a spike count difference of approximately 1 spike per 100 ms in a pair of receptors. As these differences are in the order of the statistical spike count fluctuations, the error probabilities for an ideal observer of the spike count are substantially higher (13-18%) than the errors observed in the behavioural tests (only 5%). We conclude that the animals' nervous system has to sample information from a pool of 6-13 receptors to arrive at the observed behavioural precision.