Self-organizing map (SOM) is an artificial neural network algorithm that learns from the input data through a learning function. It has been used frequently with transcriptomic data analysis, in particular for clustering co-expressed genes as a basis to infer co-regulated genes. It can be applied to any set of objects as long as a distance function can be defined between objects. SOM is numerically illustrated together with a simple UPGMA method to contrast between the two. A lesser-known application of SOM is in discovering heterogeneous motifs present in a set of sequences, making it more general than the Gibbs sampler in de novo motif discovery. These two approaches, one with a (gene × expression) matrix as input and the other with a set of sequences as input, are illustrated.