To accomplish analyses on the properties of neuronal populations it is mand
atory that each unit activity is identified within the overall noise backgr
ound and the other unit signals merged in the same trace. The problem, addr
essed as a clustering one, is particularly difficult as no assumption can b
e made on the prior data distribution. We propose an algorithm that achieve
s this goal by a two-phase agglomerative hierarchical clustering. First, an
inflated estimation (overly) of the number of clusters is cast down and, b
y a maximum entropy principle (MEP) approach, is made to collapse towards a
n arrangement near natural ones. In the second step consecutive partitions
are created by merging, two at time previously aggregated partitions, accor
ding to similarity criteria, in order to reveal a cluster solution. The pro
cedure makes no assumptions about data distributions and guarantees high ro
bustness with respect to noise. An application on real data out of multiple
unit recordings from spinal cord neurons of mixed gas-anaesthetized rats i
s presented. (C) 1998 Elsevier Science B.V. All rights reserved.