Significant evidence suggests that hydrocarbon deposits are associated with abnormally high values of seismic attenuation, and the ability to detect such zones would aid seismic exploration. Unfortunately, attenuation is difficult to measure and it is not clear how to interpret observed frequency responses. Based on forward modelling, we believe that the effect of the frequency dependent reflection coefficient which results from high dispersion in the hydrocarbon saturated zone can often be the dominant observable effect. We show how the theory of spectral decomposition can be used to detect such effects and validate the technique with synthetic data. We show examples of spectral anomalies associated with gas reservoirs in field data, and demonstrate how these anomalies can be modelled in terms of gas-induced dispersion.