When strong noise exists on local seismic traces with low signal-to-noise ratio (SNR), super-virtual interferometry (SVI) method can be used to increase the SNR of first breaks on far-offset traces, but may decrease the SNR of first breaks around the noisy traces. To solve this problem, the similarity-weighted super-virtual interferometry is developed. Correlation and convolution are applied to stack the first arrivals on neighboring traces in common phase, and consequently increase the SNR of first arrivals. The introduction of similarityweighted function improves the ability to suppress strong local abnormal noise. Both the synthetic and field data examples demonstrate the effectiveness of the proposed method to enhance seismic first breaks. At last, a discussion about the applicabilities and the anti-noise abilities of the proposed method is included.