In clastic rock, the interpretation of facies from the spectral gamma-ray log should be applied with caution because a consistent relationship with gamma-ray is lacking due to variations in provenance and/or diagenesis. In this thesis the use of the gamma-ray log for the derivation of source rock characteristics and/or diagenesis is examined. This is done by reconstructing part of the history of the sedimentary rock by modelling or minimizing the variance caused by selective transport in the gamma-ray signal and interpreting the residual signal. Two di?erent approaches were studied. In the ?rst approach a model was suggested which simulates the selective transport process. The composition of sediment subject to selective sorting is modelled as a compositional linear trend where the proportion of minerals increase or decrease as function of the settling velocity of the grains. Diagenesis and mixing of sediment sources is neglected in this model. With the forward model characteristics re?ecting the parent lithology such as composition and radio-nuclide concentration can be used to simulate a gamma-ray signature. Iterative forward modelling was used in an attempt to derive provenance characteristics from the gamma-ray signatures. The performance of the model was tested on synthetic and real gamma-ray signatures. With the iterative forward model an excellent ?t with the well log gamma-ray was found however, the resulting radio-nuclide concentrations show an unrealistic large variation. Two conclusions were drawn based on these outcomes, ?rst the model is very sensitive to the noise present in gamma-ray logs and secondly not enough constraints are available to produce realistic results from the model. The second approach is based on simplifying assumption concerning selective transport. It is assumed sediment within a su?ciently small grain-size class is deposited under similar hydraulic conditions and therefore has a comparable (chemical) composition if there are no variation in provenance or diagenesis. If this is the case, the gamma-ray log can be reconstructed from a grain size record if we take into account the variation caused by the averaging e?ect of the detector. The detectors response was approximated based on attenuation e?ects in the formation and allows us to degrade the high resolution gamma-ray log derived from the grain size record to a resolution matching the well log gamma-ray. Application to carboniferous core E10-3 resulted in a good ?t and realistic gamma-ray signatures for the grain size classes. Thorium showed the highest dependency with grain size, potassium and uranium show a comparable and less pronounced correlation. In case of core E10-3 the residual variance is expected to be mainly caused by diagenesis (formation of kaolinite), degree of sorting and organic content and it was concluded that in this case the residual signal is hard to interpret. The model was also applied to the point bar deposits of the Huesca dataset. Again the thorium content showed the highest dependency with grain size, the uranium content showed less strong correlation and no clear relationship of the grain size record with the potassium content was found. Individual processing of the point bar deposits clearly showed an increased potassium content in the point bar at the depth interval 57m - 61m. Several plausible explanations can be posed for the increased potassium content, for example the potas- sium content could be present as a solution in the pore structure, higher proportion of K-feldspar or increased radio-activity of the potassium bearing minerals.