We have demonstrated that it is possible to use NMO-based velocity analysis approach to obtain an acceptable velocity model from the very noisy simultaneous-source data. The similarity-weighted semblance can obtain a better velocity spectrum than the conventional semblance, with higher resolution and reliability. When the blending interference is so strong that the seismic reflections can not be observed, the similarity-weighted semblance can still show plausible energy peaks in the velocity spectrum, and the peaks can be picked easily. We use both simulated synthetic and field data examples to show the potential of the similarity-weighted semblance in velocity analysis of simultaneous-source data. We also compare the migrated images of unblended field data, and numerically blended field data using different picked velocities. The migrated image of blended data using the picked velocity from the similarity-weighted semblance is very close to the migrated image of unblended data, which shows great potential that the separation of simultaneous sources is no longer necessary.