Processing of raw seismic data into AVO/AVA data serves many purposes, but also induces some unwanted features (errors) in the resulting data set. Here we study the effect of such processing in an idealized case with a synthetic raw data set. The behavior of the processing errors are estimated using a statistical Gaussian model. The 1D marginal distribution of this model show a good match with observed errors. The subsequent linearized inversion reveals that the processing errors can only be safely ignored for a signal-to-noise ratio (S/N) of 0,4 or below when using an uncorrelated noise model. Such inversion results will have poor posterior resolution. Uncorrelated models with a higher S/N will be biased. Using the estimated Gaussian model to describe the noise in the data eliminates this bias and increases resolution in linear inversion. In a real-world case we expect the threshold of 0.4 to be even lower.