Computational neural networks are known to have the capability to pred
ict complex mappings between input and output data. These new tools se
em to be well-suited to NMR data. To treat simultaneously a whole set
of compounds in the alkane family, we used a back-propagation neural n
etwork with a topological description as input. The results allow for
a good prediction of the shifts because of the range of the test popul
ation (up to 62% of the known environments) and since ali types of car
bons are taken into account without distinction of connectivity.