This is a comparative theoretical study of cage effect in the photolysis of two compounds, azomethane and dimethylnitrosamine. Both compounds belong to classes, azoalkanes and nitrosamines, which undergo an almost complete inhibition of photodissociation in condensed phase. We have set up a semiquantitative simulation of the primary photochemical step, based on Langevin dynamics and surface hopping. This required the ab initio determination of quasi-diabatic states and potential energy surfaces (PES). In the PES of both compounds one can identify a region of strong nonadiabatic coupling, marked by the presence of conical intersections, which determines molecular geometry and time of the internal conversion (IC) from S-1 to S-0. Azomethane undergoes a very fast IC when the torsion angle around the N=N bond is about 90 degrees, followed by vibrational energy loss to the solvent: the outcome is isomerization rather than dissociation. On the contrary, the dissociation of nitrosamine takes place essentially in the S-1 potential surface: in this case, the mechanism of cage effect can be described as fast recombination of the fragments, favored by strong nonadiabatic coupling with S-0 at medium-long internuclear distances.