Transposable elements (TEs) are mobile, repetitive DNA sequences that are almost ubiquitous in prokaryotic and eukaryotic genomes. They have a large impact on genome structure, function and evolution. With the recent development of high-throughput sequencing methods, many genome sequences have become available, making possible comparative studies of TE dynamics at an unprecedented scale. Several methods have been proposed for the de novo identification of TEs in sequenced genomes. Most begin with the detection of genomic repeats, but the subsequent steps for defining TE families differ. High-quality TE annotations are available for the Drosophila melanogaster and Arabidopsis thaliana genome sequences, providing a solid basis for the benchmarking of such methods. We compared the performance of specific algorithms for the clustering of interspersed repeats and found that only a particular combination of algorithms detected TE families with good recovery of the reference sequences. We then applied a new procedure for reconciling the different clustering results and classifying TE sequences. The whole approach was implemented in a pipeline using the REPET package. Finally, we show that our combined approach highlights the dynamics of well defined TE families by making it possible to identify structural variations among their copies. This approach makes it possible to annotate TE families and to study their diversification in a single analysis, improving our understanding of TE dynamics at the whole-genome scale and for diverse species.