Abstract Involved in mitotic condensation, interaction of transcriptional regulatory elements or isolation of structural domains, understanding loop formation is becoming a paradigm in the deciphering of chromatin architecture and its functional role. Despite the emergence of increasingly powerful genome visualization techniques, the high variability in cell populations and the randomness of conformations still make loop detection a challenge. We introduce a new approach for determining the presence and frequency of loops in a collection of experimental conformations obtained by multiplexed super-resolution imaging. Based on a spectral approach, in conjunction with neural networks, this method offers a powerful tool to detect loops in large experimental data sets, both at the population and single cell level. The method’s performance is confirmed by applying it to recently published experimental data, where it provides a detailed and statistically quantified description of the global architecture of the chromosomal region under study.