Stokesian Dynamics has been used to investigate the origins of shear thickening in concentrated colloidal suspensions. For this study, we considered a monolayer suspension composed of charge-stabilized non-Brownian monosized rigid spheres dispersed at an areal fraction of ϕa=0.74 in a Newtonian liquid. The suspension was subjected to a linear shear field. In agreement with established experimental data, our results indicate that shear thickening in this system is associated with an order–disorder transition of the suspension microstructure. Below the critical shear rate at which this transition occurs, the suspension microstructure consists of two-dimensional analogues of experimentally observed sliding layer configurations. Above this critical shear rate, suspensions are disordered, contain particle clusters, and exhibit viscosities and microstructures characteristic of suspensions of non-Brownian hard spheres. In addition, suspensions possessing the sliding layer microstructure at the beginning of supercritical shearing tend to retain this microstructure for a period of time before disordering. The onset of this disorder is due to the formation of particle doublets within the suspension. Once formed, these doublets rotate, due to the bulk motion, and disrupt the long-range order of the suspension. The cross-stream component of the centre-to-centre separation vector associated with the two particles forming a doublet, which is zero when the doublet is perfectly aligned with the bulk velocity vector, grows exponentially with time. This strongly suggests that the evolution of these doublets is due to a change in the stability of the sliding layer configurations, with this type of ordered microstructure being linearly unstable above a critical shear rate. This contention is supported by results of a stability analysis. The analysis shows that a single string of particles is subject to a linear instability leading to the formation of particle doublets. Simulations were repeated with different numbers of particles in the computational domain, with the results found to be qualitatively independent of system size.