Abstract: In a nuclear accident, release source term is the key issue of nuclear emergency and consequences assessment. Source inversion for nuclear accidents is a way to estimate the source term by using the environmental monitoring data during the accident. Since it does not rely on the reactor state parameters, it receives a wide attention after the Fukushima nuclear accident. The modeling of source inversion for nuclear accidents based on variational data assimilation (VAR) is able to get the global optimal result. However, its result is great influenced by the error of the atmospheric dispersion model used in source inversion. In order to reduce the effect of the dispersion model error on the estimated source term, VAR coping with truncated total least squares (TTLS-VAR) was proposed. It is able to correct the dispersion model operator and monitoring data vector to enhance the accuracy of source inversion. The TTLS-VAR was verified using the date of wind tunnel experiment. The results demonstrated that the accuracy of the result calculated by the TTLS-VAR is higher than that by the VAR.