Abstract Using semi-analytic models of galaxy formation, we investigate the properties of z∼3 galaxies and compare them with the observed population of Lyman-break galaxies (LBGs). In addition to the usual quiescent mode of star formation, we introduce a physical model for starbursts triggered by galaxy–galaxy interactions. We find that with the merger rate that arises naturally in the CDM-based merging hierarchy, a significant fraction of bright galaxies identified at high redshift ( z≳2) are likely to be low-mass, bursting satellite galaxies. The abundance of LBGs as a function of redshift and the luminosity function of LBGs both appear to be in better agreement with the data when the starburst mode is included, especially when the effects of dust are considered. The objects that we identify as LBGs have observable properties including low velocity dispersions that are in good agreement with the available data. In this “Bursting Satellite” scenario, quiescent star formation at z≳2 is relatively inefficient and most of the observed LBGs are starbursts triggered by satellite mergers within massive halos. In high-resolution N-body simulations, we find that the most massive dark matter halos cluster at redshift z∼3 much as the Lyman-break galaxies (LBGs) are observed to do. This is true for both the Ω=1 CHDM model and low- Ω ΛCDM and OCDM models, all of which have fluctuation power spectra P( k) consistent with the distribution of low-redshift galaxies. The Bursting Satellite scenario can resolve the apparent paradox of LBGs that cluster like massive dark matter halos but have narrow linewidths and small stellar masses.