The challenge of physical models of musical instruments is to provide a global description of a highly complex instrument which has to be accurate but simple enough to be numerically implemented. For this purpose, a logical approach is to extract, from a theoretical or experimental study, what are the most important phenomenon involved which are thus to be taken into account. Two examples of such an approach are given by the physical models of human voice and brass instruments presented here. Although both instruments share in common the same basic principle, a self-oscillating source coupled with a resonator, crucial physical aspects are quite different. It is shown that, for brass instruments, the lips' self-sustained oscillations are mainly driven by the acoustical coupling, therefore the simpliest one degree of freedom model for the lips is sufficient. In the case of voicing, the acoustical feedback, although effective, is not sufficient to maintain self-sustained oscillations. A more elaborated vocal-folds model is needed. Regarless of the sound source, the description of the resonator is also discussed. While, for voice simulations, significant improvement can be gained using a two-dimensional linear theory, nonlinear propagation is sometimes to be taken into account for brass instruments.