We investigate the influence of uncorrelated high frequency Gaussian noise on the accuracy and reliability of first-break arrival time picks. Using a reliable automatic picking algorithm based on the Akaike information criterion (AIC) and systematically noise contaminated synthetic data, allows us to develop a characteristic relation between the mis-pick (i.e., the error in the picked arrival time) and the signal-to-noise (S/N) ratio for a specific waveform. For typical near-surface crosshole georadar survey setups, we further study the impact of such noise-related picking errors on tomographically reconstructed velocity images. Using synthetic and field data examples, results of tomographic inversions illustrate that significant velocity distortions can be introduced by noise-related mis-picking. In addition, we show that, under favorable conditions, it is possible to correct the noise-related picking errors prior to inversion.