All-optical spin switching represents a new frontier in femtomagnetism. However, its underlying principles are quite different from traditional thermal activated spin switching. Here, we employ an atomic spin model and present a systematic investigation from a single spin to a large system of over a million spins. We find that for a single spin without an external perturbation, the conservation of total angular momentum requires that the spin change, if any, exactly matches the orbital momentum change, but a laser pulse significantly alters this relation, where the spin change does not necessarily follow the orbital change. This is reflected in the strong dependence of switching on laser polarization. To have an efficient spin switching, the electron initial momentum direction must closely follow the spin's orientation, so the orbital angular momentum is transverse to the spin and consequently the spin-orbit torque lies in the same direction as the spin. The module of the spin-orbit torque is λ|S||r||P| cos 2 α + cos 2 β − 2 cos α cos β cos γ, where α(β) is the angle between spin S and position r(momentum P) and γ is the angle between r and P. These findings are manifested in a much larger system. We find that the spin response depends on underlying spin structures. A linearly polarized laser pulse creates a dip in a uniform inplane-magnetized thin film, but has little effects on Néel and Bloch walls. Both right-and left-circularly polarized light (σ + and σ − ) have stronger but different effects in both uniform spin domains and Néel walls. While σ + light creates a basin of spins pointing down, σ − light creates a mound of spins pointing up. In the vicinity of the structure spins are reversed, similar to the experimental observation. σ + light has a dramatic effect, disrupting spins in Bloch walls. By contrast, σ − light has a small effect on Bloch walls because σ − only switches down spins up and once the spins already point up, there is no major effect. These findings are expected to have important implications in the future.