We investigate the biases and uncertainties in estimates of physical parameters of high-redshift Lyman break galaxies (LBGs), such as stellar mass, mean stellar population age, and star formation rate (SFR), obtained from broad-band photometry. By combining LCDM hierarchical structure formation theory, semi-analytic treatments of baryonic physics, and stellar population synthesis models, we construct model galaxy catalogs from which we select LBGs at redshifts z ~ 3.4, 4.0, and 5.0. The broad-band spectral energy distributions (SEDs) of these model LBGs are then analysed by fitting galaxy template SEDs derived from stellar population synthesis models with smoothly declining SFRs. We compare the statistical properties of LBGs' physical parameters -- such as stellar mass, SFR, and stellar population age -- as derived from the best-fit galaxy templates with the intrinsic values from the semi-analytic model. We find some trends in these distributions: first, when the redshift is known, SED-fitting methods reproduce the input distributions of LBGs' stellar masses relatively well, with a minor tendency to underestimate the masses overall, but with substantial scatter. Second, there are large systematic biases in the distributions of best-fit SFRs and mean ages, in the sense that single-component SED-fitting methods underestimate SFRs and overestimate ages. We attribute these trends to the different star formation histories predicted by the semi-analytic models and assumed in the galaxy templates used in SED-fitting procedure, and to the fact that light from the current generation of star-formation can hide older generations of stars. These biases, which arise from the SED-fitting procedure, can significantly affect inferences about galaxy evolution from broadband photometry.