In this paper, we present a study of structural changes which may occur as a function of surface composition within a single surface reconstruction. We examine in detail the GaAs{001}(2×4) surface by measuring the ion beam desorbed Ga+ion angular distributions as a function of As coverage. The distributions of these ions, initiated by 3‐keV Ar+ion bombardment at normal incidence, are measured as a function of azimuthal and polar angles of ejection from the molecular‐beam‐epitaxy‐grown (2×4) surface. The distributions display extreme anisotropy, indicative of a specific collision sequence which results from the covalent nature of the GaAs crystal. More subtle blocking and channeling effects are also found in the angular pattern which allow qualitative analysis of the (2×4) surface structure. These results agree with the currently accepted model of missing As dimers for this reconstruction, which appear as open channels parallel to the 〈011̄〉 direction on the surface. The As coverage of the (2×4) surface was modified by molecular beam deposition of either As or Ga. These results indicate structural variations within the range of As and Ga coverages which are considered to yield the (2×4) reconstruction. The excess Ga atoms appear to adsorb in the fourfold hollow sites on top of the As dimers and at least 0.25 monolayer of Ga can be deposited before the surface reconstructs to a (3×1) symmetry. It is proposed that this adsorbed Ga may destabilize the (2×4) surface by opening the dimers, and thus initiating the reconstruction. Excess As atoms appear to adsorb in the fourfold hollow sites of the open channels on the (2×4) surface. The As coverage of the (2×4) surface was also modified by ultrahigh vacuum annealing. Angle‐resolved secondary ion distributions and reflection high‐energy electron diffraction patterns indicate that the surface reconstructs. It is concluded that the use of molecular beam deposition to vary the surface As coverage may be more effective than the thermal desorption of As2from a prepared surface.