In this paper, we compare two attenuation inversion strategies used in waveform tomography. The first strategy is a two-steps scheme: inverting velocity model and fixing attenuation model (1/Q) first, then with fixed velocity model to invert the attenuation model. Another strategy is to invert for these two models simultaneously. Although there are no big differences in synthetic data test, because the trade-off between update of the velocity and the attenuation factor is very sensitive to noise, there are big differences in real data tests. The first strategy can successfully avoid cross contamination of the velocity and attenuation fields. We also use curved layer perturbation tests to verify the high attenuation layer in the final attenuation model.