The uneven internal structures of clusters were recognized quite early on. By looking at Wolf 's plot of the galaxy distribution in Coma it is easy to spot the south-western subcluster dominated by NGC 4839 - see Fig. 18. This was re-discovered by Shane & Wirtanen  in 1954, more than half a century later. The subcluster is clearly visible in their Plates no.303 and no.1613 - here reproduced in Fig. 19 - and the authors suggested it could be a distant cluster seen in projection in the Coma cluster region. Shane & Wirtanen  classified clusters in two broad classes: regular Coma-like and irregular Virgo-like clusters. The uneven structure of the Virgo cluster had of course been noticed very early (e.g. Zwicky ). However, it is remarkable that subclustering in the prototype regular cluster was also noticed very early, but apparently ignored until being re-discovered in the X-ray . A telling example is that of Oemler . In 1976 he remarked that the giant galaxy NGC 4839 was quite an exception in his class, because there was not ``any evidence of clustering of galaxies around NGC 4839''!
Figure 18. The density of nebulae in the region of Coma, according to Wolf (1901). Note the south-western extension (north is up, east is to the left). Every grid element is 28' × 60'.
Figure 19. Contour maps of the Coma cluster of nebulæ, based on smoothed counts by 10' squares. Plate n.303 is on the left and no.1613 is on the right. From Shane & Wirtanen (1954).
The first systematic analyses of subclustering in galaxy clusters date back to the early 60's. Sydney van den Bergh [474, 476] analyzed the distribution of velocity differences among pairs of galaxies in the Virgo and Coma clusters. He compared the observed distributions to those obtained from azimuthal scramblings of the data-sets - see Fig. 20 - and found evidence for subclustering in both clusters, on ~ 0.1 Mpc scales: ``Taken at face value, this result implies that subclustering occurs in the Coma cluster.'' Abell et al.  analyzed eight clusters and found evidence for subclustering in six of them, but not in Coma. However, Abell  remarked that accounting for the presence of subclusters could not remove the mass discrepancy problem (see Section 4.2).
Figure 20. The observed distribution of velocity differences of pairs of galaxies in Virgo with separation smaller than 10', compared to the expected distribution for optical pairs. From van den Bergh (1960b).
In 1973, Bahcall  first noticed the existence of substructures around the two central dominant galaxies of Coma, NGC 4874 and NGC 4889. Her result was later confirmed by Rood , and refined, many years later, by Perea et al. , Fitchett & Webster , and Mellier et al. . Bahcall also suggested that these subclusters should be detectable as X-ray sources, independent from the cluster itself, a suggestion confirmed by Vikhlinin et al.  21 years later.
According to Dressler , another evidence for subclustering was given by the secondary peaks detected in the density profiles of several clusters [419, 335, 110].
Subclusters became theoretically appealing after White 's n-body simulations showed that ``clusters form by the progressive amalgamation of an inhomogeneous system of subclusters''.
Thanks to the increasing angular resolution of X-ray observations, subclusters started to be found also in this band. In 1979 Gorenstein et al.  attributed the granularity in the Coma cluster X-ray emission to subclustering, and a hint of the south-western subcluster could already be seen in Johnson et al. 's X-ray map of Coma. A major breakthrough came with the Einstein IPC images of Jones et al. . They showed that the X-ray morphologies of clusters, far from being smooth and spherically symmetric, were quite often irregular and clumpy. Subclustering was a common feature of galaxy clusters!
In 1982 Geller & Beers  draw density-contour maps of the galaxy distributions in 65 clusters and identified subclusters in 40 % of them. The techniques for the detection of subclusters have considerably improved in more recent years, but subsequent works have roughly confirmed this fraction [147, 155]. With gravitational lensing techniques it is now possible to look for subcondensations directly in the mass distribution, and the existence of dark subcluster has been suggested (see KNEIB, these proceedings).