The accuracy associated with assessing the environmental consequences of an accidental release of radioactivity is highly dependent on the knowledge of the source term characteristics, which are generally poorly known. The development of an automated numerical technique that integrates the radiological measurements with atmospheric dispersion modeling for more accurate source term estimation is reported. Often, this process of parameter estimation is performed by an emergency response assessor, who takes an intelligent first guess at the model parameters, then, comparing the model results with whatever measurements are available, makes an intuitive, informed next guess of the model parameters. This process may be repeated any number of times until the assessor feels that the model results are reasonable in terms of the measured observations. This process may be a most time-consuming activity that is not always suitable for real-time source term and dose assessment. Furthermore, this approach does not necessarily achieve the optimal solution because of the complicated, nonlinear relationships between the input parameters and the model predictions, because of the generally time-varying nature of source emissions and meteorology. A new approach, based on a nonlinear least-squares regression scheme coupled with the existing Atmospheric Release Advisory Capability three-dimensional atmospheric dispersion models, is to supplement the assessor’s intuition with automated mathematical methods that do not significantly increase the response time of the existing predictive models. The viability of the approach is evaluated by estimation of the known SF6 tracer release rates associated with the Mesoscale Atmospheric Transport Studies tracer experiments conducted at the Savannah River Laboratory during 1983. These 19 experiments resulted in 14 successful, separate tracer releases with sampling of the tracer plumes along the cross-plume arc situated ∼30 km from the release site. The regression technique optimizes the agreement between the measured and model-predicted tracer concentrations by varying the model input parameters within reasonable ranges of uncertainties. The technique generally estimated the measured tracer release rates within a factor of 2, with the worst estimate being within a factor of 5. This level of accuracy is quite reasonable in view of the sparse tracer concentration measurements (none closer than 30 km), the uncertainties associated with the spatial representativeness of the meteorological data, and the limitations of the models. These results, albeit for a relatively simple source term, suggest that the regression methodology is sufficiently promising to warrant further development and testing for more complex source terms.