Inverse problem of radiation-sources restoration from near-field measurement data is considered. Case of data distortion by additive noise and spikes is investigated. New approach based on a concept of “spatial extent” is suggested. It states that the weighted sum of spatial extent of solution discrepancy and spatial extent of solution should be a minimum. Thus, the concept of spatial extent is used two times, namely, in an original space (which is the space of solution) and in a conjugate space (which is the space of solution discrepancy). The model of radiation sources is described by a set of ideal Hertzian dipoles. The case, when dipoles are located along a straight line, which is parallel to measurement line of electric field, and when their electric moments are perpendicular to the measurement plane, is considered. Algorithm, based on the conjugate gradient method, is proposed. Numerical simulations of direct and inverse problems are presented.