Due to new, fast detectors, an increasing amount of information with more and more parameters has to be handled in nearly every branch of science. The extensive information is bound in multidimensional data sets and often consists of mixtures of groupable subsets and errors that might for example be explained by the instruments' or measurement uncertainties.
The Infrared Astronomical
Satellite (IRAS), started in 1983, detected around 245.000 sources.
The Deep Near Infrared Southern Sky Survey (DeNIS,
Epchtein et al. 1994),
started in 1995, will reach a number of objects which is estimated
to be around 
. Such data sets consist of different types of
objects, such as galaxies, stars or planetary nebulae.
Furthermore the instruments register multiple properties of each source
(such as coordinates, different fluxes or magnitudes). One of the main tasks is now the
exact separation of different types, by means of known properties, and
to get statistical information about objects without such collected properties.
The algorithm presented here is able to separate different structures and to give a probability function which indicates if a source is part of one of the structures.
In order to obtain assumed substructures it is possible to model the superimposing main structure with any kind of artificial form (for example multidimensional Gaussian distributions) as well as to use natural subsets (for example obtained by special information about a part of the whole data set) as a main structure. The algorithm substracts this main structure from the whole data set, the remaining part (thereafter called residuals) opens the view to eventually existing covered structures.
It is possible to improve the model of the main component by using the algorithm in an iterative way.