The Raman G-band of six single wall carbon nanotube samples with known diameters between 1.05 and 1.56 nm was measured with 14 different lasers in the 457-676 nm range. Spectra were reduced in size by discrete cosine transformation and introduced as input vectors into a Kohonen self-organizing map (SOM). Even though no diameter data was supplied to the network, the SOM was able to identify spectra belonging to the same sample. The results correlated well with those obtained by conventional cluster analysis. This suggests that artificial neural networks are capable of extracting diameter information from the G-band fine structure. (C) 2003 Elsevier B.V. All rights reserved.