Single-atom alloys, which are prepared by embedding isolated metal sites in host metals, are promising systems for improved catalyst selectivity. For technical applications, catalysts based on nanoparticles are preferred thanks to a large surface area. Herein, we investigate hydrogenation of acetylene to ethylene using kinetic Monte Carlo simulations based on density functional theory and compare the performance of Pd/Cu nanoparticles with Pd(111) and Pd/Cu(111). We find that embedding Pd in Cu systems strongly enhances the selectivity and that the reaction mechanism is fundamentally different for nano-particles and extended surfaces. The reaction mechanism on nanoparticles is complex and involves elementary steps that proceed preferentially over different sites. Edge and corner sites on nanoparticles are predicted to lower the selectivity, and we infer that a rational design strategy in selective acetylene hydrogenation is to maximize the number of (111) sites in relation to edge sites for Pd/Cu nanoparticles.