Source characterization for an unknown contaminant puff release is usually achieved by application of artificial intelligence techniques to invert a transport and dispersion model, generally by testing a carefully chosen sequence of trial inputs to the forward model for how well they permit it to reproduce observed concentrations. There are, however, meteorological situations in which the transport component dominates the spread of an airborne contaminant, allowing successful back-calculation without recourse to a sequence of forward model runs. This situation arises when the puff size is approximately that of the dominant eddies, in which case the puff is deformed by those eddies faster than it is dispersed by the smaller eddies. Two common settings in which this occurs are the convective boundary layers (for puffs with a horizontal scale similar to the boundary layer depth) and in the mid-latitude troposphere (for puffs with a horizontal scale similar to that of baroclinic cyclones. In these settings back trajectories computed for contaminated air parcels tend to converge on the source location as time approaches that of the release. Convergence is not perfect of course, both because of dispersion by smaller eddies and because the wind field in the dominant eddies will not be perfectly resolved by observations. Our algorithm, a back trajectory model for hazard origin estimation or BackHOE is tested here in the boundary layer setting, with transport times on the scale of minutes and distances on the order of a kilometer. Metrics for diagnosing maximum convergence of the cloud of contaminated parcels, thus deducing the source location, are discussed. The basic method is then extended via cluster analysis to allow diagnosis of multi-source events. The frozen wave approximation is applied in those cases where the sensor trajectories leave the observation grid. Results show that BackHOE can succeed in characterizing an unknown instantaneous source in those meteorological situations wherein puff deformation by the transporting wind field poses challenges for conventional forward-model-based AI techniques.