Using a dataset of 530 analysed mudstone samples from 19 North Sea and 9 Gulf of Mexico wells, back propagation artificial neural networks (ANNs) have been trained to estimate the clay content (proportion of particles smaller than 2 μm diameter), grain density and total organic content (TOC) of mudstones from standard wireline log data (gamma, resistivity, sonic, density, calliper). ANNs have also been trained to discriminate carbonates from clastic mudstones and also give a preliminary indication of the extent to which mudstones are lithified or cemented. Results show that for clay content, 85% of predictions are within ±10% of the measured value; for TOC, 92% of predictions are within ±1% of the measured value; for grain density, 91% of predictions are within ±0.07 g cm−3 of the measured value; for the discrimination of carbonates from clastics, 98.3% of carbonate samples and 99.9% of non-carbonate samples are classified correctly. The ANNs work well not only in the areas from where training data were measured, but also (as an example) in offshore West Africa. Potential applications of the technique include (1) the possibility to define the 3D sedimentary architecture of mudstone sequences from wireline data and, because both the porosity–effective stress and porosity–permeability relationships of mudstones are strongly influenced by clay content, (2) more accurate, basin-scale fluid flow modelling.