The aim of the paper is to compare and evaluate different approaches to detecting regular equivalence classes of valued networks. The evaluated approaches include different versions of REGE (indirect approaches) and generalized blockmodeling approaches (direct approaches). In addition to the approaches designed to detect regular equivalence, some approaches designed to detect structural equivalence are also included for comparison. The evaluation is done by means of simulations. Networks of 11 and 25 units were generated based on different known (max-) regular blockmodels and partitions. The obtained partitions were compared to the original (known) partition using the Adjusted Rand Index. The results show that homogeneity blockmodeling, implicit blockmodeling and REGE are usually the best approaches for detecting regular equivalence. The most surprising result is that methods for detecting structural equivalence preformed relatively well on networks generated based on (max)-regular equivalence, better then several approaches designed to detect regular equivalence.