Decades of geophysical monitoring have revealed the importance of slow aseismic fault slip in the release of tectonic energy. Although significant progress have been made in imaging aseismic slip on natural faults, many questions remain concerning its physical control. Here we present an attempt to study the evolution of aseismic slip in the controlled environment of the laboratory. We develop a kinematic inversion method, to image slip during the nucleation phase of a dynamic rupture within a saw‐cut sample loaded in a tri‐axial cell. We use the measurements from a strain gauge array placed in the vicinity of the fault, and the observed shortening of the sample, to invert the fault slip distribution in space and time. The inversion approach relies both on a deterministic optimization step followed by a Bayesian analysis. The Bayesian inversion is initiated with the best model reached by the deterministic step, and allows to quantify the uncertainties on the inferred slip history. We show that the nucleation consists of quasi‐static aseismic slip event expanding along the fault at a speed of the order of 200 m., before degenerating into a dynamic rupture. The total amount of aseismic slip accumulated during this nucleation phase reaches m locally, about 8%–15 % of the coseismic slip. The resolution of the method is evaluated, indicating that the main limitation is related to the impossibility of measuring strain inside the rock sample. The results obtained however show that the method could improve our understanding of earthquake nucleation.