Context: Solar-like oscillations have been observed in numerous red giants from ground and from space. An important question arises: could we expect to detect non-radial modes probing the internal structure of these stars? Aims: We investigate under what physical circumstances non-radial modes could be observable in red giants; what would be their amplitudes, lifetimes and heights in the power spectrum (PS)? Methods: Using a non-radial non-adiabatic pulsation code including a non-local time-dependent treatment of convection, we compute the theoretical lifetimes of radial and non-radial modes in several red giant models. Next, using a stochastic excitation model, we compute the amplitudes of these modes and their heights in the PS. Results: Distinct cases appear. Case A corresponds to subgiants and stars at the bottom of the ascending giant branch. Our results show that the lifetimes of the modes are mainly proportional to the inertia I, which is modulated by the mode trapping. The predicted amplitudes are lower for non-radial modes. But the height of the peaks in the PS are of the same order for radial and non-radial modes as long as they can be resolved. The resulting frequency spectrum is complex. Case B corresponds to intermediate models in the red giant branch. In these models, the radiative damping becomes high enough to destroy the non-radial modes trapped in the core. Hence, only modes trapped in the envelope have significant heights in the PS and could be observed. The resulting frequency spectrum of detectable modes is regular for ℓ=0 and 2, but a little more complex for ℓ=1 modes because of less efficient trapping. Case C corresponds to models of even higher luminosity. In these models the radiative damping of non-radial modes is even larger than in the previous case and only radial and non-radial modes completely trapped in the envelope could be observed. The frequency pattern is very regular for these stars. The comparison between the predictions for radial and non-radial modes is very different if we consider the heights in the PS instead of the amplitudes. This is important as the heights (not the amplitudes) are used as detection criterion. CIFIST Marie Curie Excellence Team.