The development of the nervous system requires nerve fibres to be guided accurately over long distances in order to make correct connections between neurons. Molecular gradients help to direct these growing fibres, by a process known as chemotaxis. However, this requires the accurate measurement of concentration differences by chemoreceptors. Here, we ask how the signals from a set of chemoreceptors interacting with a concentration gradient can best be used to determine the direction of this gradient. Prior models of chemotaxis have typically assumed that the chemoreceptors produce signals reflecting just the time-averaged binding state of those receptors. In this article, we show that in fact the optimal chemotaxis performance can be achieved when, in addition, each receptor also signals the number of unbound-to-bound transitions it experiences within the observation period. Furthermore, we show that this leads to an effective halving of the observation period required for a given level of performance. We also demonstrate that the degradation in performance observed to occur at high concentrations experimentally is likely to result not from noise intrinsic to receptor binding, but rather from noise in subsequent downstream signalling. © 2009 Elsevier B.V. All rights reserved.