Here we demonstrate the feasibility of using an array of live insects to detect concentrated packets of odor and infer the location of an odor source (∼15 m away) using a backward Lagrangian dispersion model based on the Langevin equation. Bayesian inference allows uncertainty to be quantified, which is useful for robotic planning. The electroantennogram (EAG) is the biopotential developed between the tissue at the tip of an insect antenna and its base, which is due to the massed response of the olfactory receptor neurons to an odor stimulus. The EAG signal can carry tens of bits per second of information with a rise time as short as 12 ms (K A Justice 2005 J. Neurophiol. 93 2233-9). Here, instrumentation including a GPS with a digital compass and an ultrasonic 2D anemometer has been integrated with an EAG odor detection scheme, allowing the location of an odor source to be estimated by collecting data at several downwind locations. Bayesian inference in conjunction with a Lagrangian dispersion model, taking into account detection errors, has been implemented resulting in an estimate of the odor source location within 0.2 m of the actual location.