A new method of clustering for forward-tracking algorithms has been developed to reconstruct the tracks of gamma-rays in high-resolution detector systems such as AGATA (Advanced GAmma Tracking Array). This technique, called Deterministic Annealing Filter (DAF), comes from statistical physics and is used in high-energy physics. After a description of the DAF method and of the forward-tracking algorithm, the performance of this clustering method is discussed in terms of photopeak efficiency and peak-to-total ratio obtained with GEANT4 simulations for the AGATA geometry. A comparison with the standard so-called "cone clustering method" shows similar performances with a better photopeak efficiency at high multiplicity for the DAF algorithm. (C) 2010 Elsevier B.V. All rights reserved.