Using a convergent beam approach, a spatially resolved ellipsometry (SRE) has been achieved. The modifications and the limitations due to the use of such a nonplanar wave for an ellipsometric measurement are discussed. We first established the relations between the ellipsometric data (tan ψ,cos Δ), the optical properties of the sample which are assumed to be spatially nonuniform [rp(x,y) and rs(x,y)], and the optical characteristics of the beam (i.e., electric field distribution of the light on the sample surface and its angular domain Fourier transform function). These relations are established for both a coherent and an incoherent light source. We use them to investigate the absolute accuracy achievable with SRE on an homogeneous sample. It is demonstrated that submonolayer sensitivity can be obtained using a 10×10‐μm spot. In the case of a step discontinuity of rp(x,y) and rs(x,y), strong optical resonances can take place in SRE. Their quantitative analysis is performed in the (tan ψ,cos Δ) plane. In this representation, the optical resonance corresponds to a trajectory, called ‘‘spatial trajectory.’’ Another type of trajectory can be observed for a multilayer structure in the case when one (or more) thickness is slowly varying versus the position (‘‘thickness trajectory’’). The combination of both cases leads to a generalized trajectory concept which allows for a quantitative analysis of two‐dimensional ellipsometric maps. This trajectory approach is illustrated on two practical situations: (i) a SiO2 layer etched through a lithographic mask on top of a GaAs wafer; (ii) metallic patterns deposited on GaAs. It is shown that the trajectory concept is capable of analyzing patterns with lateral dimensions smaller than the actual spot size (i.e., direct optical r- esolution of SRE).