Summary: Real-world networks show community structures – groups of nodes that are densely intra-connected and sparsely inter-connected to other groups. Nevertheless, Community Detection (CD) is non-trivial, since identifying these groups of nodes according to their local connectivity can hold many plausible solutions, leading to the creation of different methods. In particular, CD has recently been achieved by Force-Directed Algorithms (FDAs), which originally were designed as a way to visualize networks. FDAs map the network nodes as particles in a \(D\)-dimensional space that are affected by forces acting in accordance to the connectivity. However, the literature on FDA-based methods for CD has grown in parallel from the classical methods, leaving several open questions, such as how accurately FDAs can recover communities compared to classical methods. In this work, we start to fill these gaps by evaluating different numerical implementations of 5 FDA methods and different clustering analyses on state-of-the-art network benchmarks – including networks with or without weights and networks with a hierarchical organisation. We also compare these results with 8, well-known, classical CD methods. Our findings show that FDA methods can achieve higher accuracy than classical methods, albeit their effectiveness depends on the chosen setting – with optimisation techniques leading over numerical integration and distance-based clustering algorithms leading over density-based ones. Overall, our work provides detailed information for any researcher aiming to apply FDAs for community detection.